Cisco HyperFlex All-NVMe Systems for Deploying Microsoft SQL Server 2019 Databases with VMware ESXi

Deployment Guide for Microsoft SQL Server 2019 Databases on Cisco HyperFlex All-NVMe Systems with VMware ESXi 6.7 Update 2

NOTE: Works with document’s Advanced Properties “First Published” property. Click File | Properties | Advanced Properties | Custom.

Last Updated: January 14, 2020

NOTE: Works with document’s Advanced Properties “Last Updated” property. Click File | Properties | Advanced Properties | Custom.

                                                    

About the Cisco Validated Design Program

The Cisco Validated Design (CVD) program consists of systems and solutions designed, tested, and documented to facilitate faster, more reliable, and more predictable customer deployments. For more information, go to:

http://www.cisco.com/go/designzone.

ALL DESIGNS, SPECIFICATIONS, STATEMENTS, INFORMATION, AND RECOMMENDATIONS (COLLECTIVELY, "DESIGNS") IN THIS MANUAL ARE PRESENTED "AS IS," WITH ALL FAULTS.  CISCO AND ITS SUPPLIERS DISCLAIM ALL WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE WARRANTY OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT OR ARISING FROM A COURSE OF DEALING, USAGE, OR TRADE PRACTICE.  IN NO EVENT SHALL CISCO OR ITS SUPPLIERS BE LIABLE FOR ANY INDIRECT, SPECIAL, CONSEQUENTIAL, OR INCIDENTAL DAMAGES, INCLUDING, WITHOUT LIMITATION, LOST PROFITS OR LOSS OR DAMAGE TO DATA ARISING OUT OF THE USE OR INABILITY TO USE THE DESIGNS, EVEN IF CISCO OR ITS SUPPLIERS HAVE BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGES.

THE DESIGNS ARE SUBJECT TO CHANGE WITHOUT NOTICE.  USERS ARE SOLELY RESPONSIBLE FOR THEIR APPLICATION OF THE DESIGNS.  THE DESIGNS DO NOT CONSTITUTE THE TECHNICAL OR OTHER PROFESSIONAL ADVICE OF CISCO, ITS SUPPLIERS OR PARTNERS.  USERS SHOULD CONSULT THEIR OWN TECHNICAL ADVISORS BEFORE IMPLEMENTING THE DESIGNS.  RESULTS MAY VARY DEPENDING ON FACTORS NOT TESTED BY CISCO.

CCDE, CCENT, Cisco Eos, Cisco Lumin, Cisco Nexus, Cisco StadiumVision, Cisco TelePresence, Cisco WebEx, the Cisco logo, DCE, and Welcome to the Human Network are trademarks; Changing the Way We Work, Live, Play, and Learn and Cisco Store are service marks; and Access Registrar, Aironet, AsyncOS, Bringing the Meeting To You, Catalyst, CCDA, CCDP, CCIE, CCIP, CCNA, CCNP, CCSP, CCVP, Cisco, the Cisco Certified Internetwork Expert logo, Cisco IOS, Cisco Press, Cisco Systems, Cisco Systems Capital, the Cisco Systems logo, Cisco Unified Computing System (Cisco UCS), Cisco UCS B-Series Blade Servers, Cisco UCS C-Series Rack Servers, Cisco UCS S-Series Storage Servers, Cisco UCS Manager, Cisco UCS Management Software, Cisco Unified Fabric, Cisco Application Centric Infrastructure, Cisco Nexus 9000 Series, Cisco Nexus 7000 Series. Cisco Prime Data Center Network Manager, Cisco NX-OS Software, Cisco MDS Series, Cisco Unity, Collaboration Without Limitation, EtherFast, EtherSwitch, Event Center, Fast Step, Follow Me Browsing, FormShare, GigaDrive, HomeLink, Internet Quotient, IOS, iPhone, iQuick Study,  LightStream, Linksys, MediaTone, MeetingPlace, MeetingPlace Chime Sound, MGX, Networkers, Networking Academy, Network Registrar, PCNow, PIX, PowerPanels, ProConnect, ScriptShare, SenderBase, SMARTnet, Spectrum Expert, StackWise, The Fastest Way to Increase Your Internet Quotient, TransPath, WebEx, and the WebEx logo are registered trademarks of Cisco Systems, Inc. and/or its affiliates in the United States and certain other countries.

All other trademarks mentioned in this document or website are the property of their respective owners. The use of the word partner does not imply a partnership relationship between Cisco and any other company. (0809R)

© 2020 Cisco Systems, Inc. All rights reserved.

 

Table of Contents

Executive Summary. 4

Solution Overview.. 5

Technology Overview.. 6

Solution Design. 19

Microsoft SQL Server Deployment 31

Solution Resiliency Testing and Validation. 39

Database Performance Testing. 42

Common Database Maintenance Scenarios. 47

Troubleshooting Performance. 48

Summary. 49

About the Author 50

 

 

Cisco HyperFlex™ Systems deliver complete hyperconvergence, combining software-defined networking and computing with the next-generation Cisco HyperFlex Data Platform. Engineered on the Cisco Unified Computing System™ (Cisco UCS®), Cisco HyperFlex Systems deliver the operational requirements for agility, scalability, and pay-as-you-grow economics of the cloud—with the benefits of on-premises infrastructure. Cisco HyperFlex Systems deliver a pre-integrated cluster with a unified pool of resources that you can quickly deploy, adapt, scale, and manage to efficiently power your applications and your business.

With the HyperFlex 4.0. release, Cisco introduced NVMe (Non-Volatile Memory Express) drive option in its HyperFlex All-NVMe platform where in the cache, capacity and housekeeping SATA/SAS drives are replaced by NVMe drives. The system is co-engineered with Intel VMD for HotPlug and surprise removal access of NVMe drives and at the same time achieving high performance without compromising Intel RAS (Reliability, Availability and Serviceability) capabilities. With the help of high speed PCIe bus interface and its protocol efficiencies, NVMe drives provide much better IO performance than SAS/SATA drives. With support for NVMe drives for storage, high compute capacity powered by the latest Intel Cascade lake CPUs, 25/40GbE converged network bandwidth and with its continued enhancements in the log structured Distributed File System, HyperFlex All-NVMe systems brings high performing hyperconverged platform supporting various mission critical database workloads which demands lower latency and consistent performance.

With the latest All-NVMe storage configurations, a low latency, high performing hyperconverged storage platform has become a reality. This makes the storage platform optimal to host the latency sensitive applications like Microsoft SQL Server. This document provides the considerations and deployment guidelines to have a Microsoft SQL server virtual machine setup on an All-NVMe Cisco HyperFlex Storage Platform.

Introduction

Cisco HyperFlex™ Systems unlock the potential of hyperconvergence. The systems are based on an end-to-end software-defined infrastructure, combining software-defined computing in the form of Cisco Unified Computing System (Cisco UCS) servers; software-defined storage with the powerful Cisco HX Data Platform and software-defined networking with the Cisco UCS fabric. Together with a single point of connectivity and hardware management, these technologies deliver a pre-integrated and an adaptable cluster that is ready to provide a unified pool of resources to power applications as your business needs dictate.

Microsoft SQL Server 2019 is the latest relational database engine release from Microsoft. It brings in a lot of new features and enhancements to the relational and analytical engines. It is built to provide a consistent and reliable database experience to applications delivering high performance. Currently more and more database deployments are getting virtualized and hyperconverged storage solutions are gaining popularity in the enterprise space. Cisco HyperFlex All-NVMe system is the latest hyperconverged storage solution which was released recently. It provides a high performing and cost-effective storage solution making use of the NVME drives locally attached to the VMware ESXi hosts. It is crucial to understand the best practices and implementation guidelines that enable customers to run a consistently high performing SQL server database solution on a hyperconverged All-NVMe solution.

Audience

This document can be referenced by system administrators, database specialists and storage architects, who work on planning, designing and implementing Microsoft SQL Server database solution on Cisco HyperFlex Hyperconverged systems.

Purpose of this Document

This document discusses reference architecture and implementation guidelines for deployment of SQL Server database instances on Cisco HyperFlex All-NVMe solution.

What’s New?

The following are highlights of the solution discussed in this CVD:

·       Microsoft SQL Server 2019 deployment and validation on the latest HyperFlex All-NVMe platform.

·       Support for latest intel 2^nd generation Cascade Lake CPUs, high performing Intel Optane NVMe drives for cache and Intel P4500 high Performance and endurance value NVMe drives for storage capacity.

 

This section contains the following:

·       HyperFlexDataPlatform40

·       Architecture

·       Physical Infrastructure

·       Cisco HyperFlex Systems Details

·       HyperFlex All-NVMe Systems for Database Deployments

HyperFlex Data Platform 4.0 – All-NVMe Storage Platform

Cisco HyperFlex Systems are designed with an end-to-end software-defined infrastructure that eliminates the compromises found in first-generation products. Cisco HyperFlex Systems combine software-defined computing in the form of Cisco UCS® servers, software-defined storage with the powerful Cisco HyperFlex HX Data Platform Software, and software-defined networking (SDN) with the Cisco® unified fabric that integrates smoothly with Cisco Application Centric Infrastructure (Cisco ACI™). With All-NVMe memory storage configurations, and a choice of management tools, Cisco HyperFlex Systems deliver a pre-integrated cluster that is up and running in an hour or less and that scales resources independently to closely match your application resource needs (Figure 1).

Figure 1        Cisco HyperFlex Systems Offer Next-Generation Hyperconverged Solutions

The Cisco HyperFlex All-NVMe HX Data Platform includes:

·       Enterprise-class data management features that are required for complete lifecycle management and enhanced data protection in distributed storage environments—including replication, always on inline deduplication, always on inline compression, thin provisioning, instantaneous space efficient clones, and snapshots.

·       Simplified data management that integrates storage functions into existing management tools, allowing instant provisioning, cloning, and pointer-based snapshots of applications for dramatically simplified daily operations.

·       Improved control with advanced automation and orchestration capabilities and robust reporting and analytics features that deliver improved visibility and insight into IT operation.

·       Independent scaling of the computing and capacity tiers, giving you the flexibility to scale out the environment based on evolving business needs for predictable, pay-as-you-grow efficiency. As you add resources, data is automatically rebalanced across the cluster, without disruption, to take advantage of the new resources.

·       Continuous data optimization with inline data deduplication and compression that increases resource utilization with more headroom for data scaling.

·       Dynamic data placement optimizes performance and resilience by making it possible for all cluster resources to participate in I/O responsiveness. All-NVMe nodes use NVMe drives for caching layer as well as capacity layer. This approach helps eliminate storage hotspots and makes the performance capabilities of the cluster available to every virtual machine. If a drive fails, reconstruction can proceed quickly as the aggregate bandwidth of the remaining components in the cluster can be used to access data.

·       Compute-only nodes, powered by the latest Intel generations of CPUs, provide enormous compute resources required by performance sensitive applications like databases.

·       Low latency and lossless 25G and 40G unified Ethernet networking backed by Cisco UCS 6400 and 6300 Series Fabric Interconnects which increase the reliability, efficiency, and scalability of Ethernet networks.

·       Enterprise data protection with a highly-available, self-healing architecture that supports non-disruptive, rolling upgrades and offers call-home and onsite 24x7 support options.

·       API-based data platform architecture that provides data virtualization flexibility to support existing and new cloud-native data types.

·       Cisco Intersight is the latest visionary cloud-based management tool, designed to provide a centralized off-site management, monitoring and reporting tool for all your Cisco UCS based solutions including HyperFlex Cluster.

Architecture

In Cisco HyperFlex Systems, the data platform spans three or more Cisco HyperFlex HX-Series nodes to create a highly available cluster. Each node includes a Cisco HyperFlex HX Data Platform controller that implements the scale-out and distributed file system using internal NVMe SSD drives to store data. The controllers communicate with each other over 25 or 40 Gigabit Ethernet to present a single pool of storage that spans the nodes in the cluster (Figure 2). Nodes access data through a data layer using file, block, object, and API plug-ins. As nodes are added, the cluster scales linearly to deliver computing, storage capacity, and I/O performance.

Figure 2        Distributed Cisco HyperFlex System

In the VMware vSphere environment, the controller occupies a virtual machine with a dedicated number of processor cores and amount of memory, allowing it to deliver consistent performance and not affect the performance of the other virtual machines on the cluster. The controller can access all storage without hypervisor intervention through the VMware VM_DIRECT_PATH feature. In the All-NVMe-memory configuration, the controller uses the node’s memory, a dedicated NVMe drive for write logging, and other NVMe drives for distributed capacity storage. The controller integrates the data platform into VMware software using two preinstalled VMware ESXi vSphere Installation Bundles (VIBs):

·       IO Visor: This VIB provides a network file system (NFS) mount point so that the VMware ESXi hypervisor can access the virtual disk drives that are attached to individual virtual machines. From the hypervisor’s perspective, it is simply attached to a network file system.

·       VMware Storage API for Array Integration (VAAI): This storage offload API allows VMWare vSphere to request advanced file system operations such as snapshots and cloning. The controller causes these operations to occur through manipulation of metadata rather than actual data copying, providing rapid response, and thus rapid deployment of new application environments.

Physical Infrastructure

Cisco Unified Computing System

The Cisco Unified Computing System (Cisco UCS) is a next-generation data center platform that unites compute, network and storage access. The platform, optimized for virtual environments, is designed using open industry-standard technologies and aims to reduce the total cost of ownership (TCO) and increase the business agility. The system integrates a low-latency; lossless 10 or 40 Gigabit Ethernet unified network fabric with enterprise-class, x86-architecture servers. It is an integrated, scalable, multi-chassis platform in which all resources participate in a unified management domain.

Cisco Unified Computing System consists of the following components:

·       Compute - The system is based on an entirely new class of computing system that incorporates rack mount and blade servers based on Intel® Xeon® scalable processors product family.

·       Network - The system is integrated onto a low-latency, lossless, 25 or 40-Gbps unified network fabric. This network foundation consolidates Local Area Networks (LAN’s), Storage Area Networks (SANs), and high-performance computing networks which are separate networks today. The unified fabric lowers costs by reducing the number of network adapters, switches, and cables, and by decreasing the power and cooling requirements.

·       Virtualization - The system unleashes the full potential of virtualization by enhancing the scalability, performance, and operational control of virtual environments. Cisco security, policy enforcement, and diagnostic features are now extended into virtualized environments to better support changing business and IT requirements.

·       Storage access - The system provides consolidated access to both SAN storage and Network Attached Storage (NAS) over the unified fabric. It is also an ideal system for Software Defined Storage (SDS). Combining the benefits of single framework to manage both the compute and Storage servers in a single pane, Quality of Service (QOS) can be implemented if needed to inject IO throttling in the system. In addition, the server administrators can pre-assign storage-access policies to storage resources, for simplified storage connectivity and management leading to increased productivity. In addition to external storage, both rack and blade servers have internal storage which can be accessed through built-in hardware RAID controllers. With storage profile and disk configuration policy configured in Cisco UCS Manager, storage needs for the host OS and application data gets fulfilled by user defined RAID groups for high availability and better performance.

·       Management - the system uniquely integrates all system components to enable the entire solution to be managed as a single entity by Cisco UCS Manager (UCSM). Cisco UCS Manager has an intuitive graphical user interface (GUI), a command-line interface (CLI), and a powerful scripting library module for Microsoft PowerShell built on a robust application programming interface (API) to manage all system configuration and operations.

Cisco Unified Computing System is designed to deliver:

·       Reduced Total Cost of Ownership and increased business agility.

·       Increased IT staff productivity through just-in-time provisioning and mobility support.

·       A cohesive, integrated system which unifies the technology in the data center. The system is managed and tested.

·       Scalability through a design for hundreds of discrete servers and thousands of virtual machines and the capability to scale I/O bandwidth to match the demand.

·       Industry standard supported by a partner ecosystem of industry leaders.

Cisco UCS Fabric Interconnect

The Cisco UCS Fabric Interconnect (FI) is a core part of Cisco Unified Computing System, providing both network connectivity and management capabilities for the system.  Both Cisco UCS 6400 and 6300 Series Fabric Interconnects are supported. Depending on the model chosen, the Cisco UCS Fabric Interconnect offers line-rate, low-latency, lossless 25 Gigabit or 40 Gigabit Ethernet, Fibre Channel over Ethernet (FCoE) and Fibre Channel connectivity. Cisco UCS Fabric Interconnects provide the management and communication backbone for the Cisco UCS C-Series, S-Series, and HX-Series Rack-Mount Servers, Cisco UCS B-Series Blade Servers and Cisco UCS 5100 Series Blade Server Chassis. All servers and chassis, and therefore all blades, attached to the Cisco UCS Fabric Interconnects become part of a single, highly available management domain. In addition, by supporting unified fabrics, the Cisco UCS Fabric Interconnects provide both the LAN and SAN connectivity for all servers within its domain.

The Cisco UCS 6454 54-Port Fabric Interconnect is a One-Rack-Unit (1RU) 10/25/40/100 Gigabit Ethernet, FCoE and Fibre Channel switch offering up to 3.82 Tbps throughput and up to 54 ports. The switch has 36 10/25-Gbps Ethernet ports, 4 1/10/25-Gbps Ethernet ports, 6 40/100-Gbps Ethernet uplink ports and 8 unified ports that can support 8 10/25-Gbps Ethernet ports or 8/16/32-Gbps Fibre Channel ports. All Ethernet ports are capable of supporting FCoE.

The Cisco UCS 6332-16UP Fabric Interconnect is a one-rack-unit (1RU) 10/40 Gigabit Ethernet, FCoE, and native Fibre Channel switch offering up to 2430 Gbps of throughput. The switch has 24 40-Gbps fixed Ethernet and FCoE ports, plus 16 1/10-Gbps fixed Ethernet, FCoE, or 4/8/16 Gbps FC ports. Up to 18 of the 40-Gbps ports can be reconfigured as 4x10Gbps breakout ports, providing up to 88 total 10-Gbps ports, although Cisco HyperFlex nodes must use a 40GbE VIC adapter in order to connect to a Cisco UCS 6300 Series Fabric Interconnect.

The Cisco UCS 6332 Fabric Interconnect is a one-rack-unit (1RU) 40 Gigabit Ethernet and FCoE switch offering up to 2560 Gbps of throughput. The switch has 32 40-Gbps fixed Ethernet and FCoE ports. Up to 24 of the ports can be reconfigured as 4x10Gbps breakout ports, providing up to 96 10-Gbps ports, although Cisco HyperFlex nodes must use a 40GbE VIC adapter in order to connect to a Cisco UCS 6300 Series Fabric Interconnect.

Cisco HyperFlex HX-Series Nodes

A standard HyperFlex cluster requires a minimum of three HX-Series “converged” nodes (with disk storage). Data is replicated across at least two of these nodes, and a third node is required for continuous operation in the event of a single-node failure. Each node that has disk storage is equipped with at least one high-performance NVMe or SSD drive for data caching and rapid acknowledgment of write requests. Each node also is equipped with additional disks, up to the platform’s physical limit, for long term storage and capacity.

Cisco HyperFlex HXAF220c-M5N All-NVMe Node

This small footprint Cisco HyperFlex all-NVMe model contains a 240 GB M.2 form factor solid-state disk (SSD) that acts as the boot drive, a 1 TB housekeeping NVMe SSD drive, a single 375 GB Intel Optane NVMe SSD write-log drive, and six to eight 1 TB or 4 TB NVMe SSD drives for storage capacity. Optionally, the Cisco HyperFlex Acceleration Engine card can be added to improve write performance and compression. Self-encrypting drives are not available as an option for the all-NVMe nodes.

Figure 3        HXAF220c-M5N All-NVMe node

Cisco VIC 1457 and 1387 mLOM Interface Cards

The Cisco UCS VIC 1457 is a quad-port Small Form-Factor Pluggable (SFP28) mLOM card designed for the M5 generation of Cisco UCS C-Series Rack Servers. The card supports 10-Gbps or 25-Gbps Ethernet and FCoE, where the speed of the link is determined by the model of SFP optics or cables used. The card can be configured to use a pair of single links, or optionally to use all four links as a pair of bonded links. The Cisco UCS VIC 1457 is used in conjunction with the Cisco UCS 6454 model Fabric Interconnect.

The Cisco UCS VIC 1387 Card is a dual-port Enhanced Quad Small Form-Factor Pluggable (QSFP+) 40-Gbps Ethernet and Fibre Channel over Ethernet (FCoE)-capable PCI Express (PCIe) modular LAN-on-motherboard (mLOM) adapter installed in the Cisco UCS HX-Series Rack Servers. The Cisco UCS VIC 1387 is used in conjunction with the Cisco UCS 6332 or 6332-16UP model Fabric Interconnects.

The mLOM slot can be used to install a Cisco VIC without consuming a PCIe slot, which provides greater I/O expandability. It incorporates next-generation converged network adapter (CNA) technology from Cisco, providing investment protection for future feature releases. The card enables a policy-based, stateless, agile server infrastructure that can present up to 256 PCIe standards-compliant interfaces to the host, each dynamically configured as either a network interface card (NICs) or host bus adapter (HBA). The personality of the interfaces is set programmatically using the service profile associated with the server. The number, type (NIC or HBA), identity (MAC address and World-Wide Name [WWN]), failover policy, adapter settings, bandwidth, and quality-of-service (QoS) policies of the PCIe interfaces are all specified using the service profile.

Cisco HyperFlex Compute-Only Nodes

All current model Cisco UCS M4 and M5 generation servers, except the Cisco UCS C880 M4 and Cisco UCS C880 M5, may be used as compute-only nodes connected to a Cisco HyperFlex cluster, along with a limited number of previous M3 generation servers. Any valid CPU and memory configuration is allowed in the compute-only nodes, and the servers can be configured to boot from SAN, local disks, or internal SD cards. The following servers may be used as compute-only nodes:

·       Cisco UCS B200 M5 Blade Server

·       Cisco UCS B480 M5 Blade Server

·       Cisco UCS C220 M5 Rack Mount Servers

·       Cisco UCS C240 M5 Rack Mount Servers

·       Cisco UCS C480 M5 Rack Mount Servers

Cisco HyperFlex Systems Details

Engineered on the successful Cisco UCS platform, Cisco HyperFlex Systems deliver a hyperconverged solution that truly integrates all components in the data center infrastructure—compute, storage, and networking. The HX Data Platform starts with three or more nodes to form a highly available cluster. Each of these nodes has a software controller called the Cisco HyperFlex Controller. It takes control of the internal locally installed drives to store persistent data into a single distributed, multitier, object-based data store. The controllers communicate with each other over low-latency 25 or 40 Gigabit Ethernet fabric, to present a single pool of storage that spans across all the nodes in the cluster so that data availability is not affected if single or multiple components fail.

Data Distribution

Incoming data is distributed across all nodes in the cluster to optimize performance using the caching tier (Figure 4). Effective data distribution is achieved by mapping incoming data to stripe units that are stored evenly across all nodes, with the number of data replicas determined by the policies you set. When an application writes data, the data is sent to the appropriate node based on the stripe unit, which includes the relevant block of information. This data distribution approach in combination with the capability to have multiple streams writing at the same time avoids both network and storage hot spots, delivers the same I/O performance regardless of virtual machine location, and gives you more flexibility in workload placement. This contrasts with other architectures that use a data locality approach that does not fully use available networking and I/O resources and is vulnerable to hot spots.

Figure 4        Data is Striped Across Nodes in the Cluster

 

When moving a virtual machine to a new location using tools such as VMware Dynamic Resource Scheduling (DRS), the Cisco HyperFlex HX Data Platform does not require data to be moved. This approach significantly reduces the impact and cost of moving virtual machines among systems.

Data Operations

The data platform implements a distributed, log-structured file system that changes how it handles caching and storage capacity depending on the node configuration.

In the All-NVMe configuration, the data platform uses a caching layer in NVMe disk to accelerate write responses, and it implements the capacity layer in NVMe as well. Read requests are fulfilled directly from data obtained from the NVMe drives in the capacity layer. A dedicated read cache is not required to accelerate read operations.

Incoming data is striped across the number of nodes required to satisfy availability requirements—usually two or three nodes. Based on policies you set, incoming write operations are acknowledged as persistent after they are replicated to the NVMe drives in other nodes in the cluster. This approach reduces the likelihood of data loss due to NVMe disk or node failures. The write operations are then de-staged to NVMe disks in the capacity layer in the All-NVMe configuration for long-term storage.

The log-structured file system writes sequentially to one of two write logs (three in case of RF=3) until it is full. It then switches to the other write log while de-staging data from the first to the capacity tier.  When existing data is (logically) overwritten, the log-structured approach simply appends a new block and updates the metadata. This layout benefits NVMe configurations in which seek operations are not time consuming. It reduces the write amplification levels of NVMe disk and the total number of writes the flash media experiences due to incoming writes and random overwrite operations of the data.

When data is de-staged to the capacity tier in each node, the data is deduplicated and compressed. This process occurs after the write operation is acknowledged, so no performance penalty is incurred for these operations. A small deduplication block size helps increase the deduplication rate. Compression further reduces the data footprint. Data is then moved to the capacity tier as write cache segments are released for reuse (Figure 5).

Figure 5        Data Write Operation Flow Through the Cisco HyperFlex HX Data Platform

Hot data sets—data that are frequently or recently read from the capacity tier are cached in memory. Unlike Hybrid configurations, All-NVMe and All-Flash configurations do not use an NVMe or SSD read cache since there are no performance benefits using a cache; the persistent data copy already resides on the high-performance NVMe or SSDs. In these configurations, a read cache implemented with NVMe or SSDs could become a bottleneck and prevent the system from using the aggregate bandwidth of the entire set of NVMe or SSDs.

Data Optimization

The Cisco HyperFlex HX Data Platform provides a finely detailed inline deduplication and variable block inline compression that is always on for objects in the cache (NVMe and memory) and capacity (NVMe) layers. Unlike other solutions, which require you to turn off these features to maintain performance, the deduplication and compression capabilities in the Cisco data platform are designed to sustain and enhance performance and significantly reduce physical storage capacity requirements.

Data Deduplication

Data deduplication is used on all storage in the cluster, including memory and NVMe drives. Based on a patent-pending Top-K Majority algorithm, the platform uses conclusions from empirical research that show that most data, when sliced into small data blocks, has significant deduplication potential based on a minority of the data blocks. By fingerprinting and indexing just these frequently used blocks, high rates of deduplication can be achieved with only a small amount of memory, which is a high-value resource in cluster nodes (Figure 6).

Figure 6        Cisco HyperFlex HX Data Platform Optimizes Data Storage with No Performance Impact

 

Inline Compression

The Cisco HyperFlex HX Data Platform uses high-performance inline compression on data sets to save storage capacity. Although other products offer compression capabilities, many negatively affect performance. In contrast, the Cisco data platform uses CPU-offload instructions to reduce the performance impact of compression operations. In addition, the log-structured distributed-objects layer has no effect on modifications (write operations) to previously compressed data. Instead, incoming modifications are compressed and written to a new location, and the existing (old) data is marked for deletion, unless the data needs to be retained in a snapshot.

The data that is modified does not need to be read prior to the write operation. This feature avoids typical read-modify-write penalties and significantly improves write performance.

Log-Structured Distributed Objects

In the Cisco HyperFlex HX Data Platform, the log-structured distributed-object store layer groups and compresses data that filters through the deduplication engine into self-addressable objects. These objects are written to disk in a log-structured, sequential manner. All incoming I/O—including random I/O—is written sequentially to both the caching (NVMe and memory) and persistent tiers. The objects are distributed across all nodes in the cluster to make uniform use of storage capacity.

By using a sequential layout, the platform helps increase flash-memory endurance. Because read-modify-write operations are not used, there is little or no performance impact of compression, snapshot operations, and cloning on overall performance.

Data blocks are compressed into objects and sequentially laid out in fixed-size segments, which in turn are sequentially laid out in a log-structured manner (Figure 7). Each compressed object in the log-structured segment is uniquely addressable using a key, with each key fingerprinted and stored with a checksum to provide high levels of data integrity. In addition, the chronological writing of objects helps the platform quickly recover from media or node failures by rewriting only the data that came into the system after it was truncated due to a failure.

Figure 7        Cisco HyperFlex HX Data Platform Optimizes Data Storage with No Performance Impact

Encryption

Securely encrypted storage optionally encrypts both the caching and persistent layers of the data platform. Integrated with enterprise key management software, or with passphrase-protected keys, encrypting data at rest helps you comply with HIPAA, PCI-DSS, FISMA, and SOX regulations. The platform itself is hardened to Federal Information Processing Standard (FIPS) 140-1 and the encrypted drives with key management comply with the FIPS 140-2 standard.

Data Services

The Cisco HyperFlex HX Data Platform provides a scalable implementation of space-efficient data services, including thin provisioning, space reclamation, pointer-based snapshots, and clones—without affecting performance.

Thin Provisioning

The platform makes efficient use of storage by eliminating the need to forecast, purchase, and install disk capacity that may remain unused for a long time. Virtual data containers can present any amount of logical space to applications, whereas the amount of physical storage space that is needed is determined by the data that is written. You can expand storage on existing nodes and expand your cluster by adding more storage-intensive nodes as your business requirements dictate, eliminating the need to purchase large amounts of storage before you need it.

Snapshots

The Cisco HyperFlex HX Data Platform uses metadata-based, zero-copy snapshots to facilitate backup operations and remote replication: critical capabilities in enterprises that require always-on data availability. Space-efficient snapshots allow you to perform frequent online backups of data without needing to worry about the consumption of physical storage capacity. Data can be moved offline or restored from these snapshots instantaneously.

·       Fast snapshot updates: When modified-data is contained in a snapshot, it is written to a new location, and the metadata is updated, without the need for read-modify-write operations.

·       Rapid snapshot deletions: You can quickly delete snapshots. The platform simply deletes a small amount of metadata that is located on an SSD, rather than performing a long consolidation process as needed by solutions that use a delta-disk technique.

·       Highly specific snapshots: With the Cisco HyperFlex HX Data Platform, you can take snapshots on an individual file basis. In virtual environments, these files map to drives in a virtual machine. This flexible specificity allows you to apply different snapshot policies on different virtual machines.

Full featured backup applications, such as Veeam Backup and Replication, can limit the amount of throughput the backup application can consume which can protect latency sensitive applications during the production hours. With the release of v9.5 update 2, Veeam is the first partner to integrate HX native snapshots into the product. HX Native snapshots do not suffer the performance penalty of delta-disk snapshots, and do not require heavy disk IO impacting consolidation during snapshot deletion.

Particularly important for SQL administrators is the Veeam Explorer for SQL: https://www.veeam.com/microsoft-sql-server-explorer.html, which can provide transaction level recovery within the Microsoft VSS framework. The three ways Veeam Explorer for SQL Server works to restore SQL Server databases include; from the backup restore point, from a log replay to a point in time, and from a log replay to a specific transaction – all without taking the VM or SQL Server offline.

Fast, Space-Efficient Clones

In the Cisco HyperFlex HX Data Platform, clones are writable snapshots that can be used to rapidly provision items such as virtual desktops and applications for test and development environments. These fast, space-efficient clones rapidly replicate storage volumes so that virtual machines can be replicated through just metadata operations, with actual data copying performed only for write operations. With this approach, hundreds of clones can be created and deleted in minutes. Compared to full-copy methods, this approach can save a significant amount of time, increase IT agility, and improve IT productivity.

Clones are deduplicated when they are created. When clones start diverging from one another, data that is common between them is shared, with only unique data occupying new storage space. The deduplication engine eliminates data duplicates in the diverged clones to further reduce the clone’s storage footprint.

Data Replication and Availability

In the Cisco HyperFlex HX Data Platform, the log-structured distributed-object layer replicates incoming data, improving data availability. Based on policies that you set, data that is written to the write cache is synchronously replicated to one or two other NVMe drives located in different nodes before the write operation is acknowledged to the application. This approach allows incoming writes to be acknowledged quickly while protecting data from NVMe or node failures. If an NVMe or node fails, the replica is quickly re-created on other NVMe drives or nodes using the available copies of the data. 

The log-structured distributed-object layer also replicates data that is moved from the write cache to the capacity layer. This replicated data is likewise protected from SSD or node failures. With two replicas, or a total of three data copies, the cluster can survive uncorrelated failures of two SSD drives or two nodes without the risk of data loss. Uncorrelated failures are failures that occur on different physical nodes. Failures that occur on the same node affect the same copy of data and are treated as a single failure. For example, if one disk in a node fails and subsequently another disk on the same node fails, these correlated failures count as one failure in the system. In this case, the cluster could withstand another uncorrelated failure on a different node. See the Cisco HyperFlex HX Data Platform system administrator’s guide for a complete list of fault-tolerant configurations and settings.

If a problem occurs in the Cisco HyperFlex HX controller software, data requests from the applications residing in that node are automatically routed to other controllers in the cluster. This same capability can be used to upgrade or perform maintenance on the controller software on a rolling basis without affecting the availability of the cluster or data. This self-healing capability is one of the reasons that the Cisco HyperFlex HX Data Platform is well suited for production applications.

In addition, native replication transfers consistent cluster data to local or remote clusters. With native replication, you can snapshot and store point-in-time copies of your environment in local or remote environments for backup and disaster recovery purposes.

HyperFlex VM Replication

HyperFlex Replication copies the virtual machine’s snapshots from one Cisco HyperFlex cluster to another Cisco HyperFlex cluster to facilitate recovery of protected virtual machines from a cluster or site failure, via failover to the secondary site.

Data Rebalancing

A distributed file system requires a robust data rebalancing capability. In the Cisco HyperFlex HX Data Platform, no overhead is associated with metadata access, and rebalancing is extremely efficient. Rebalancing is a non-disruptive online process that occurs in both the caching and persistent layers, and data is moved at a fine level of specificity to improve the use of storage capacity. The platform automatically rebalances existing data when nodes and drives are added or removed or when they fail. When a new node is added to the cluster, its capacity and performance is made available to new and existing data. The rebalancing engine distributes existing data to the new node and helps ensure that all nodes in the cluster are used uniformly from capacity and performance perspectives. If a node fails or is removed from the cluster, the rebalancing engine rebuilds and distributes copies of the data from the failed or removed node to available nodes in the clusters.

Online Upgrades

Cisco HyperFlex HX-Series systems and the HX Data Platform support online upgrades so that you can expand and update your environment without business disruption. You can easily expand your physical resources; add processing capacity; and download and install BIOS, driver, hypervisor, firmware, and Cisco UCS Manager updates, enhancements, and bug fixes.

HyperFlex All-NVMe Systems for Database Deployments

SQL server database systems act as the backend to many critical and performance hungry applications. It is very important to ensure that it delivers consistent performance with predictable latency throughout. The following are some of the major advantages of Cisco HyperFlex All-NVMe hyperconverged systems which makes it ideally suited for SQL Server database implementations:

·       Low latency with consistent performance: Cisco HyperFlex All-NVMe nodes provides excellent platform for critical database deployment by offering low latency, consistent performance and exceeds most of the database service level agreements.

·       Data protection (fast clones and snapshots, replication factor, VM replication and Stretched Cluster): The HyperFlex systems are engineered with robust data protection techniques that enable quick backup and recovery of the applications in case of any failures.

·       Storage optimization: All the data that comes in the HyperFlex systems are by default optimized using inline deduplication and data compression techniques. Additionally, the HX Data Platform’s log-structured file system ensures data blocks are written to flash devices in a sequential manner thereby increasing flash-memory endurance. HX System makes efficient use of flash storage by using Thin Provisioning storage optimization technique.

·       Performance and Capacity Online Scalability: The flexible and independent scalability of the capacity and compute tiers of HyperFlex systems provide immense opportunities to adapt to the growing performance demands without any application disruption.

·       No Performance Hotspots: The distributed architecture of HyperFlex Data Platform ensures that every VM can leverage the storage IOPS and capacity of the entire cluster, irrespective of the physical node it is residing on. This is especially important for SQL Server VMs as they frequently need higher performance to handle bursts of application or user activity.

·       Non-disruptive System maintenance: Cisco HyperFlex Systems enables distributed computing and storage environment which enables the administrators to perform system maintenance tasks without disruption.

This section contains the following:

·       Introduction

·       Logical Network Design

·       Storage Configuration for SQL Guest VMs

·       Deployment Planning

Introduction

This section details the architectural components of Cisco HyperFlex, a hyperconverged system to host Microsoft SQL Server databases in a virtual environment. Figure 8 depicts a sample Cisco HyperFlex hyperconverged reference architecture comprising HX-Series All-NVMe data nodes.

Figure 8        Cisco HyperFlex Reference Architecture using All-NVMe Nodes

Cisco HyperFlex is composed of a pair of Cisco UCS Fabric Interconnects along with up to sixteen HX-Series All-NVMe data nodes per cluster. Up to 16 compute-only servers can also be added per HyperFlex cluster. Adding Cisco UCS rack mount servers and/or Cisco UCS 5108 Blade chassis, which house Cisco UCS blade servers allows for additional compute resources in an extended cluster design. Up to eight separate HX clusters can be installed under a single pair of Fabric Interconnects. The two Fabric Interconnects connect to every HX-Series rack mount server, and connect to every Cisco UCS 5108 blade chassis, and Cisco UCS rack mount server. Upstream network connections, also referred as “north bound” network, are made from the Fabric Interconnects to the customer datacenter network at the time of installation. In the above reference diagram, a pair of Cisco Nexus 9000 series switches are used and configured as vPC pairs for high availability. For more details on physical connectivity of HX-Series services, compute-only servers, Fabric Interconnect to the north bound network, please refer to the Physical Topology section of the Cisco HyperFlex 4.0 for Virtual Server Infrastructure with VMware ESXi CVD.

Infrastructure services such as Active Directory, DNS, NTP and VMWare vCenter are typically installed outside the HyperFlex cluster. Customers can leverage these existing services deploying and managing the HyperFlex cluster.

The HyperFlex storage solution has several data protection techniques, as explained in detail in the Technology overview section, one of which is data replication which needs to be configured on HyperFlex cluster creation. Based on the specific performance and data protection requirements, customer can choose either a replication factor of two (RF2) or three (RF3). For the solution validation (described in the “Solution Testing and Validation” later in this document), we had configured the test HyperFlex cluster to be of replication factor 3 (RF3).

As described in the earlier Technology Overview section, Cisco HyperFlex distributed file system software runs inside a controller VM, which gets installed on each cluster node. These controller VMs pool and manage all the storage devices and exposes the underlying storage as NFS mount points to the VMware ESXi hypervisors. The ESXi hypervisors exposes these NFS mount points as datastores to the guest virtual machines to store their data.

Logical Network Design

In the Cisco HyperFlex All-NVMe system, Cisco VIC 1387 is used to provide the required logical network interfaces on each host in the cluster. The communication pathways in the Cisco HyperFlex system can be categorized in to four different traffic zones as described below.

·       Management Zone: This zone comprises the connections needed to manage the physical hardware, the hypervisor hosts, and the storage platform controller virtual machines (SCVM). These interfaces and IP addresses need to be available to all staff who will administer the HX system, throughout the LAN/WAN. This zone must provide access to Domain Name System (DNS) and Network Time Protocol (NTP) services and allow Secure Shell (SSH) communication. In this zone are multiple physical and virtual components:

-      Fabric Interconnect management ports.

-      Cisco UCS external management interfaces used by the servers, which answer via the FI management ports.

-      ESXi host management interfaces.

-      Storage Controller VM management interfaces.

-      A roaming HX cluster management interface.

-      Storage Controller VM Management interfaces.

·       VM Zone: This zone is comprised of the connections needed to service network IO to the guest VMs that will run inside the HyperFlex hyperconverged system. This zone typically contains multiple VLANs that are trunked to the Cisco UCS Fabric Interconnects via the network uplinks and tagged with 802.1Q VLAN IDs. These interfaces and IP addresses need to be available to all staff and other computer endpoints which need to communicate with the guest VMs in the HX system, throughout the LAN/WAN.

·       Storage Zone: This zone comprises the connections used by the Cisco HX Data Platform software, ESXi hosts, and the storage controller VMs to service the HX Distributed Data Filesystem. These interfaces and IP addresses always need to be able to communicate with each other for proper operation. During normal operation, this traffic all occurs within the Cisco UCS domain, however there are hardware failure scenarios where this traffic would need to traverse the network northbound of the Cisco UCS domain. For that reason, the VLAN used for HX storage traffic must be able to traverse the network uplinks from the Cisco UCS domain, reaching FI A from FI B, and vice-versa. This zone is primarily jumbo frame traffic therefore jumbo frames must be enabled on the Cisco UCS uplinks. In this zone are multiple components:

-      A teamed interface is used for storage traffic on each Hyper-V host in the HX cluster.

-      Storage Controller VM storage interfaces.

-      A roaming HX cluster storage interface.

·       vMotion Zone: This zone comprises the connections used by the ESXi hosts to enable live migration of the guest VMs from host to host. During normal operation, this traffic all occurs within the Cisco UCS domain, however there are hardware failure scenarios where this traffic would need to traverse the network northbound of the Cisco UCS domain. For that reason, the VLAN used for HX live migration traffic must be able to traverse the network uplinks from the Cisco UCS domain, reaching FI A from FI B, and vice-versa.

By leveraging Cisco UCS vNIC templates, LAN connectivity policies and vNIC placement policies in service profile, eight vNICs are carved out from Cisco VIC 1387 on each HX-Series server for network traffic zones mentioned above. Every HX-Series server will detect the network interfaces in the same order, and they will always be connected to the same VLANs via the same network fabrics. Table 1  lists the vNICs and other configuration details used in the solution.

Table 1    Virtual Interface Order with in HX-Series Server

 

Figure 9 illustrates the logical network design of a HX-Series server of HyperFlex cluster.

Figure 9        HX-Series Server Logical Network Diagram

As shown in Figure 9, four virtual standard switches are configured for four traffic zones. Each virtual switch is configured with two vNICs and are connected to both the Fabric Interconnects. The vNICs are configured in active and standby fashion for Storage, Management and vMotion networks. However, for VM network virtual switch vNICs are configured in active and active fashion. This ensures that the data path for guest VMs traffic has aggregated bandwidth for the specific traffic type.

Jumbo frames are enabled for:

·       Storage traffic: Enabling jumbo frames on the Storage traffic zone would benefit in the following SQL server database use case scenarios:

-      Heavy write SQL server guest VMs caused by the activities such as database restoring, rebuilding indexes, importing data and so on.

-      Heavy read SQL server guest VMs caused by the typical maintenance activities such as backup database, export data, report queries, rebuilding indexes and so on.

·       vMotion traffic: Enabling jumbo frames on vMotion traffic zone help the system quickly failover the SQL VMs to other hosts; there by, reducing the overall database downtime.

Creating a separate logical network (using two dedicated vNICs) for guest VMs is beneficial with the following advantages: 

·       Isolating guest VM traffic from other traffic such as management, HX replication and so on.

·       A dedicated MAC pool can be assigned to each vNIC, which would simplify troubleshooting the connectivity issues.

·       As shown in Figure 9, the VM Network switch is configured with two vNICs in active and active fashion to provide two active data paths which will result in aggregated bandwidth.

For more details on the network configuration of the HyperFlex HX-Server node, using Cisco UCS network policies, templates and service profiles, refer to the Cisco UCS Design section in the Cisco HyperFlex 4.0 for Virtual Server Infrastructure with VMware ESXi CVD.

The following sections provide more details on configuration and deployment best practices to deploy the SQL server databases on HyperFlex All-NVMe nodes.

Storage Configuration for SQL Guest VMs

Figure 10 illustrates the storage configuration recommendations for virtual machines running SQL server databases on HyperFlex All-NVMe nodes. Single LSI Logic virtual SCSI controller is used to host the Guest OS. Separate Paravirtual SCSI (PVSCSI) controllers are configured to host SQL server data and log files. For large scale and high performing SQL deployments, it is recommended to spread the SQL data files across two or more different PVSCSI controllers for better performance as shown in the figure. Additional performance guidelines are detailed in the Deployment Planning section.

Figure 10     Storage Design for Microsoft SQL Server Database Deployment

 

Deployment Planning

It is crucial to follow and implement the configuration best practices and recommendations in order to achieve best performance from any underlying system. This section details the major design and configuration best practices that should be followed when deploying SQL server databases on All-NVMe HyperFlex systems.

Datastore Recommendation

The following recommendations can be followed while deploying the SQL server virtual machines on HyperFlex All-NVMe Systems.

All the virtual machine’s virtual disks comprising guest Operating System, SQL data, and transaction log files can be placed on a single datastore exposed as NFS file share to the ESXi hosts. Deploying multiple SQL virtual machines using single datastore simplifies the management tasks.

There is a maximum queue depth limit of 1024 for each NFS datastore per host, which is an optimum queue depth for most of the workloads. However, when consolidated IO requests from all the virtual machines deployed on the datastore exceeds 1024 (per host limit), then virtual machines might experience higher IO latencies. Symptoms of higher latencies can be identified using ESXTOP results.

In such cases, creating new datastore and deploying SQL virtual machines on the new datastore will help. The general recommendation is to deploy low IO demanding SQL virtual machines in one single datastore until high guest latencies are noticed. Also, deploying a dedicated datastore for High IO demanding SQL VMs will allow dedicated queue and hence lesser latencies can be observed.

The following figure shows that two different datastores are used for deploying various SQL guest virtual machines. “SQL-DS1” is used to deploy multiple small to medium SQL virtual machines while “SQL-DS2” is dedicatedly used for deploying single large SQL virtual machine with high IO demanding performance requirements.

Figure 11     HyperFlex Datastores

SQL Virtual Machine Configuration Recommendation

While creating a VM for deploying SQL Server instance on a HyperFlex All-NVMe system, the following recommendations should be followed for performance and better administration.

Cores per Socket

NUMA is becoming increasingly more important to ensure workloads, like databases, allocate and consume memory within the same physical NUMA node that the vCPUs are scheduled. By changing appropriate Cores per Socket, make sure the virtual machine is configured such that both memory and cpu resources can be met by single physical NUMA. In case of wide virtual machines (demanding more resources than a single physical NUMA), resources can be allocated from two or more physical NUMA groups.  For more details on virtual machine configurations best practices with varying resource requirements, please refer to this VMware KB article: https://www.vmware.com/content/dam/digitalmarketing/vmware/en/pdf/solutions/sql-server-on-vmware-best-practices-guide.pdf

Memory Reservation

SQL server database transactions are usually CPU and memory intensive. In a heavy OLTP database systems, it is recommended to reserve all the memory assigned to the SQL virtual machines. This ensures that the assigned memory to the SQL VM is committed and will eliminate the possibility of ballooning and swapping the memory out by the ESXi hypervisor. Memory reservations will have little overhead on the ESXi system. For more information about memory overhead, see Understanding Memory Overhead: https://pubs.vmware.com/vsphere-51/index.jsp?topic=%2Fcom.vmware.vsphere.resmgmt.doc%2FGUID-4954A03F-E1F4-46C7-A3E7-947D30269E34.html

Figure 12     Memory Reservations for SQL Virtual Machine

Paravirtual SCSI adapters for Large-Scale High IO Virtual Machines

For virtual machines with high disk IO requirements, it is recommended to use Paravirtual SCSI (PVSCSI) adapters. PVSCSI controller is a virtualization aware, high-performance SCSI adapter that allows the lowest possible latency and highest throughput with the lowest CPU overhead. It also has higher queue depth limits compared to other legacy controllers. Legacy controllers (LSI Logic SAS, LSI Logic Parallel and so on) can cause bottleneck and impact database performance; hence not recommended for IO intensive database applications such as SQL server databases.

Queue Depth and SCSI Controller Recommendations

Many times, queue depth settings of virtual disks are overlooked, which can impact performance particularly in high IO workloads. Systems such as Microsoft SQL Server databases tend to issue a lot of simultaneous IOs resulting in an insufficient VM driver queue depth settings (default setting is 64 for PVSCSI) to sustain the heavy IOs. It is recommended to change the default queue depth setting to a higher value (up to 254) as suggested in this VMware KB article: https://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=2053145

For large-scale and high IO databases, it is recommended to use multiple virtual disks and have those virtual disks distributed across multiple SCSI controller adapters rather than assigning all of them to a single SCSI controller. This ensures that the guest VM will access multiple virtual SCSI controllers (four SCSI controllers maximum per guest VM), which in turn results in greater concurrency by utilizing the multiple queues available for the SCSI controllers.

Virtual Machine Network Adapter type

It is highly recommended to configure virtual machine network adapters with “VMXNET 3”. VMXNET 3 is the latest generation of para-virtualized NICs designed for performance. It offers several advanced features including multi-queue support, receive side scaling, IPv4/IPv6 offloads, and MSI/MSI-X interrupt delivery. While creating a new virtual machine, choose “VMXNET 3” as the adapter type as shown in Figure 13.

Figure 13     Virtual Machine Network Adapter Type

Guest Power Scheme Settings

HX Servers are optimally configured, at factory installation time, with appropriate BIOS policy settings at the host level and hence does not require any changes. Similarly, ESXi power management option (at vCenter level) is set to “High performance” at the time of HX installation by installer as shown in Figure 14.

Figure 14     Power Setting on HX hypervisor Node

Inside the SQL server guest, it is recommended to set the power management option to “High Performance” for optimal database performance as shown in Figure 15. Starting with Windows 2019, the setting High performance is chosen by default.

Figure 15     SQL Guest VM Power Settings in Windows Operating System

For other SQL server specific configuration recommendations on virtualized environments, see SQL Server best practices guide on VMware vSphere.

Achieving Database High Availability

Cisco HyperFlex storage systems incorporates efficient storage level availability techniques such as data mirroring (Replication Factor 2/3), native snapshot etc., to make sure continuous data access to the guest VMs hosted on the cluster. More details of the HX Data Platform Cluster Tolerated Failures are detailed here: https://www.cisco.com/c/en/us/td/docs/hyperconverged_systems/HyperFlex_HX_DataPlatformSoftware/AdminGuide/3_5/b_HyperFlexSystems_AdministrationGuide_3_5/b_HyperFlexSystems_AdministrationGuide_3_5_chapter_00.html#id_13113.

This section describes the high availability techniques that are helpful to enhance the availability of the virtualized SQL server databases (apart from the storage level availability, which comes with HyperFlex solutions).

The availability of the individual SQL Server database instance and virtual machines can be enhanced using the technologies listed below:

·       VMware HA: to achieve virtual machine availability

·       Microsoft SQL Server AlwaysOn: To achieve database level high availability

Single VM / SQL Instance Level High Availability using VMware vSphere HA Feature

Cisco HyperFlex solution leverages VMware clustering to provide availability to the hosted virtual machines. Since the exposed NFS storage is mounted on all the hosts in the cluster, they act as a shared storage environment to help migrate VMs between the hosts. This configuration helps migrate the VMs seamlessly in case of planned as well as unplanned outage. The vMotion vNIC need to be configured with Jumbo frames for faster guest VM migration.

You can find more information in this VMware document: https://docs.vmware.com/en/VMware-vSphere/6.5/vsphere-esxi-vcenter-server-65-availability-guide.pdf

Database Level High Availability using SQL AlwaysOn Availability Group Feature

HyperFlex architecture inherently uses NFS datastores. Microsoft SQL Server Failover Cluster Instance (FCI) needs shared storage which cannot be on NFS storage (unsupported by VMware ESXi). Hence FCI is not supported as high availability option, instead SQL Server AlwaysOn Availability Group feature can be used. Introduced in Microsoft SQL Server 2012, AlwaysOn Availability Groups maximizes the availability of a set of user databases for an enterprise. An availability group supports a failover environment for a discrete set of user databases, known as availability databases, that failover together. An availability group supports a set of read-write primary databases and one to eight sets of corresponding secondary databases. Optionally, secondary databases can be made available for read-only access and/or some backup operations. More information on this feature can be found at the Microsoft MSDN here: https://msdn.microsoft.com/en-us/library/hh510230.aspx.

Microsoft SQL Server AlwaysOn Availability Groups take advantage of Windows Server Failover Clustering (WSFC) as a platform technology. WSFC uses a quorum-based approach to monitor the overall cluster health and maximize node-level fault tolerance. The AlwaysOn Availability Groups will get configured as WSFC cluster resources and the availability of the same will depend on the underlying WSFC quorum modes and voting configuration explained here: https://docs.microsoft.com/en-us/sql/sql-server/failover-clusters/windows/wsfc-quorum-modes-and-voting-configuration-sql-server.

Using AlwaysOn Availability Groups with synchronous replication, supporting automatic failover capabilities, enterprises will be able to achieve seamless database availability across the database replicas configured. The following figure depicts the scenario where an AlwaysOn availability group is configured between the SQL server instances running on two separate HyperFlex Storage systems. To ensure that the involved databases provide guaranteed high performance and no data loss in the event of failure, proper planning need to be done to maintain a low latency replication network link between the clusters.

Figure 16     Synchronous AlwaysOn Configuration Across HyperFlex All-NVMe Systems

Although there are no definitive rules on the infrastructure used for hosting a secondary replica, the following are some of the guidelines if you plan to have a primary replica on the All-NVMe High Performing cluster:

·       In case of a synchronous replication (no data loss)

-      The replicas need to be hosted on similar hardware configurations to ensure that the database performance is not compromised while waiting for the acknowledgment from the replicas.

-      Ensure a high-speed, low latency network connection between the replicas.

·       In case of an asynchronous replication (may have data loss)

-      The performance of the primary replica does not depend on the secondary replica, so it can be hosted on low cost hardware solutions as well.

-      The amount to data loss depends on the network characteristics and the performance of the replicas. 

If you are willing to deploy AlwaysOn Availability Group within a single HyperFlex All-NVMe cluster, which involves more than 2 replicas, VMWare DRS anti-affinity rules must be used to ensure that each SQL VM replica is placed on different VMware ESXi hosts in order to reduce database downtime. For more details on configuring VMware anti-affinity rules, see: http://pubs.vmware.com/vsphere-60/index.jsp?topic=%2Fcom.vmware.vsphere.resmgmt.doc%2FGUID-7297C302-378F-4AF2-9BD6-6EDB1E0A850A.html.

This section contains the following:

·       Cisco HyperFlex 4.0 All-NVMe System Installation and Deployment

·       Deployment Procedure

Cisco HyperFlex 4.0 All-NVMe System Installation and Deployment

This CVD focuses on Microsoft SQL Server virtual machine deployment and assumes the availability of an already running healthy All-NVMe HyperFlex 4.0 cluster. For more information on deployment of Cisco HyperFlex 4.0cluster, see: https://www.cisco.com/c/en/us/td/docs/unified_computing/ucs/UCS_CVDs/hx_4_vsi_vmware_esxi.html

Deployment Procedure

This section provides a step-by-step deployment procedure of setting up a test Microsoft SQL server 2019 using a Windows Server 2019 virtual machine on a Cisco HyperFlex All-NVMe system. Cisco recommends following the guidelines mentioned here: http://www.vmware.com/content/dam/digitalmarketing/vmware/en/pdf/solutions/sql-server-on-vmware-best-practices-guide.pdf to have an optimally performing SQL server database configuration.

To deploy, follow these steps:

1.     Before creating a guest VM and installing SQL server on the guest, it is important to gather the following information. It is assumed that information such as IP addresses, Server names, DNS/ NTP/ VLAN details of HyperFlex Systems are available before proceeding with SQL VM deployment on HX All-NVMe System. An example of the database checklist is shown in the following table.

Table 2    Virtual Interface Order with in HX-Series Server

2.     Verify HyperFlex Cluster System is healthy and configured correctly. To verify, follow these steps:

a.     Log into HX Connect dashboard using the HyperFlex Cluster IP address on a browser as shown below.

Figure 17     HyperFlex Cluster Health Status

b.     Make sure that the VMware ESXi Host service profiles in Cisco UCS Manager are all healthy without any errors. The following screenshot shows the service profile status summary from the Cisco UCS Manager UI.

Figure 18     Cisco UCS Manager Service Profile

3.     Create datastores for deploying SQL guest VMs and make sure the datastores are mounted on all the HX cluster nodes. The procedure for adding datastores to the HyperFlex system is given in the HX Administration guide. The following figure shows the creation of a sample datastore. Block size of 8K is chosen for datastore creation as it is appropriate for SQL server database.

Figure 19     HyperFlex Datastore Creation

 

4.     As described in the Deployment Procedure section of this guide, make sure that the OS, data and log files are segregated and balanced by configuring separate Paravirtual virtual SCSI controllers as shown in Figure 20. In VMware vCenter, go to Hosts and Clusters -> datacenter -> cluster -> VM-> VM properties -> Edit Settings to change the VM configuration as shown in Figure 21.

Figure 20      Sample SQL Server Virtual Machine Disk Layout

Figure 21     SQL Server Virtual Machine Configuration

5.     Mount Windows 2019 OS DVD to the Virtual Machine and install the Windows server 2019.

6.     Initialize, format, and label the volumes for Windows OS files, SQL server data and log files. Use 64K for the allocation unit size when formatting the volumes. Figure 22 (disk management utility of Windows OS) shows a sample logical volume layout of our test virtual machine.

Figure 22     SQL Server Virtual Machine Disk Layout

7.     Increase PVSCSI adapter’s queue depth by adding a registry entry inside the guest VM as described in the VMware knowledgebase article: https://kb.vmware.com/s/article/2053145. Both RequestRingPages and MaxQueueDepth should be increased to 32 and 254 respectively. Since the queue depth setting is per SCSI controller, consider additional PVSCSI controllers to increase the total number of outstanding IOPS the VM can sustain.

8.     When the Windows Guest Operating System is installed in the virtual machine, it is highly recommended to install VMware tools as explained here: https://kb.vmware.com/s/article/1014294

9.     Install Microsoft SQL Server 2019 on the Windows Server 2019 virtual machine. To install the database engine on the guest VM, refer to this Microsoft document: https://docs.microsoft.com/en-us/sql/database-engine/install-windows/install-sql-server-database-engine.

a.     Download and mount the required edition of Microsoft SQL Server 2019 ISO to virtual machine from the vCenter GUI. The choice of Standard or Enterprise edition of Microsoft SQL Server 2019 can be selected based on the application requirements.

b.     On the Server Configuration window of SQL server installation, make sure that instant file initialization is enabled by enabling check box as shown in Figure 23. This enables the SQL server data files are instantly initialized avowing zeroing operations.

Figure 23     Enabling Instant File Initialization During SQL Server Deployment

c.     If the domain account which is used as SQL server service account is not member of local administrator group, then add SQL server service account to the “Perform volume maintenance tasks” policy using Local Security Policy editor as in Figure 24.

Figure 24     Granting Volume Maintenance Task Permissions to the SQL Server Service Account

d.     In the Database Engine Configuration window under the TempDB tab, make sure the number of TempDB data files are equal to 8 when the vCPUs or logical processors of the SQL VM is less than or equal to 8. If the number of logical processors is more than 8, start with 8 data files and try to add data files in the multiple of 4 when the contention is noticed on the TempDB resources (using SQL Dynamic Management Views). The following diagram shows that there are 8 TempDB files chosen for a SQL virtual machine which has 8 vCPUs. Also, as a best practice, keep the TempDB data and log files on two different volumes.

Figure 25     TempDB Data and Log files Location

e.     Once the SQL server is successfully installed, use SQL server Configuration manager to verify that the SQL server service is up and running as shown below.

Figure 26     SQL Server Configuration Manager

 

f.      Create a user database using SQL Server Management studio or Transact-SQL so that the database logical file layout is in line with the desired volume layout. Detailed instructions are here: https://docs.microsoft.com/en-us/sql/relational-databases/databases/create-a-database

This section contains the following:

·       Node Failure Test

·       Fabric Interconnect Failure Test

·       Database Maintenance Tests

This section details some of the tests conducted to validate the robustness of the solution. These tests were conducted on a HyperFlex cluster built with four HXAF220c-M5N All-NVMe nodes. Table 3  lists the component details of the test setup.

Table 3    Hardware and Software Component Details Used in HyperFlex All-NVMe Testing and Validation

 

 The major tests conducted on the setup are as follows and will be described in detail in this section:

·       Node failure Test

·       Fabric Interconnect Failure Test

·       Database Maintenance Tests

Note that in all the tests listed in Table 3,  the HammerDB tool (www.hammerdb.com) is used to generate the required stress on the guest SQL VM.  A separate client machine located outside the HyperFlex cluster is used to run the testing tool and generate the database workload.

Node Failure Test

The intention of this failure test is to analyze how the HyperFlex system behaves when failure is introduced into the cluster on an active node (running multiple guest VMs). The expectation is that the Cisco HyperFlex system should be able to detect the failure, initiate automatic VM migration from a failed node to the one of the surviving nodes and retain the pre-failure state with an acceptable limit of performance degradation.

In our testing, node failure was introduced when the cluster is stressed with four SQL VMs (one VM per host) utilizing 60-65% of cluster IO capacity and 40% of cluster CPU utilization. When one node was powered off (unplug both power cables), SQL guest VM running on the failed node successfully failed over to one of the surviving nodes. After SQL VM migrated to the surviving node, database consistency checks were run on the database and there were no database consistency errors were reported. Later, database workload was manually restarted on the VM. The impact observed on the overall performance (IOPS dip) because of failed node was around 10-15% which is due to the absence of one node. Later when the failed node was powered up, it rejoined the cluster automatically and started syncing up with the cluster. The cluster returned to the pre-failure performance within 5-10 minutes after the failed node was brought online (including the cluster sync-up time).

Fabric Interconnect Failure Test

The intention of this failure test is to analyze the performance impact in case of a fabric interconnect switch failure. Since Cisco UCS Fabric Interconnects (FIs) are always deployed in pairs and operating as a single cluster, failure of single Fabric Interconnect should not impact systems connected to them.

In our tests, we introduced the Fabric Interconnect failure when the cluster was stressed with SQL VMs utilizing 60-65% of cluster IO capacity and 40% of cluster CPU utilization. When one of the FIs was powered off (unplugged both power supplies), no impact was observed on the VMs running the workload. All the VMs remained fully functional and no VM migrations were observed in the cluster.

Database Maintenance Tests

In any hyperconverged systems, it is important to assess the impact caused by database maintenance tasks to the regular SQL workloads running on the same cluster. The maintenance activities typically have sequential IO pattern as opposed to random IO pattern of regular OLTP workloads. Usually in typical hyperconverged shared storage environments, caution must be exercised while running DB maintenance activities during the business hours as they may impact the regular operational workloads.

The following maintenance activities were carried out on a SQL guest VM deployed on an All-NVMe cluster to assess the impact on the ongoing workload in the cluster. The cluster setup used is the same as list in Table 4.    

·       Full database restore from an external backup source (backup source residing outside of HX cluster)

·       Full database consistency check and complete backup to HX volume with in the virtual machine

·       Rebuilding indexes of two large tables (of size around 50GB)

·       Exporting SQL data to flat files (located within HX cluster)

·       Importing data from flat files into SQL (files located within HX Cluster)

Note that these maintenance activities are run on a separate SQL VM in parallel to the regular SQL VMs, which were used to exert stress on the cluster up to its 70-75% cluster IO capacity. As expected, full database restore caused 20 to 25% IOPS drop in IOPS which is understandable given the full database restore activity (100% sequential writes activity) done on a cluster which is already exercised to 70-75% resource usage by ongoing OLTP workload. Full database backup did not have any impact on the ongoing workloads as HX All-NVMe cluster could accommodate the additional read bandwidth requirement of full database backup. DB rebuild activity which is typically involves sequential reads and writes, has an impact of 15-20% on the ongoing workloads.

The amount of impact caused by the maintenance activities would typically depend on the replication factor and percentage of cluster resource utilization in addition to factors such as back up settings etc. On system with appropriate resource headroom (which is done by right capacity planning), the impact would be much lower. Below figure (HX performance dashboard GUI) shows the cluster behavior when the maintenance activities such as restore, backup and index rebuild are performed on a VM and when an operational workload is running on the other VMs in the same cluster.

Figure 27     Performance Impact Analysis of Typical Database Maintenance Activities for Ongoing Database Workload

This section contains the following:

·       Single Large VM Performance

·       Performance Scaling with Multiple VMs

·       All-NVMe Cluster for High Performance Demanding SQL Workloads

This section contains examples of different ways in which Microsoft SQL server workloads can take advantage of the HyperFlex Data Platform architecture and its performance scaling attributes.

Single Large VM Performance

The HyperFlex Data Platform uses a distributed architecture. One of the main advantages of this approach is that the cluster resources form a single, seamless pool of storage capacity and performance resources. This allows any individual VM to take advantage of the overall cluster resources and is not limited to the resources on the local node that hosts the VM. This is a unique capability and significant architectural differentiator that Cisco HyperFlex Data Platform provides.

There are a couple of deployment scenarios that are common in data centers which benefit from the Cisco HyperFlex Data Platform:

·       VM Hotspot – rarely do all the VMs in any shared virtual infrastructure show uniform utilization of resources. The capacity growth for the individual VMs usually are different and their performance requirements are different at different points in time. With the HyperFlex distributed architecture, those hotspots are easily absorbed by the infrastructure without having capacity or performance hotspots in the infrastructure.

·       Large VMs – Since the cluster presents a common pool of resources – it makes it possible to deploy large applications and VMs with performance and capacity requirements that exceed the capability of any single node in the cluster. The fact that Cisco HyperFlex supports compute only nodes to be a part of the cluster further strengthens this use case.

This section describes the attribute of the architecture through a performance test with a large SQL VM. The cluster setup used is the same as detailed in Table 3.  The details of the VM configuration and the workload are provided in Table 4.  OLTP workload with 70:30 read write ratio was exerted on the guest VM. The workload stressed the VM up to 65-70% of CPU utilization which resulted in around 40% of ESXi host CPU utilization.

Table 4    VM and Workload Details Used for Single Large VM Test

Figure 28 shows the performance of the single VM running a large SQL workload for about 3 hours. There are a few noteworthy points listed below:

·       Large VM with a very large working set size can get sustained high IOPS leveraging resources (capacity and performance) from all 4 nodes in the cluster. Note that it is possible to scale to higher IOPS with an even larger VM size.

·       Dedupe and Compression is turned on by default and the default setting is being used in this test as well.

·       Delivers consistent performance throughout the test without any drop in the performance.

Figure 28     Single Large Working Set SQL Server Database Test on HyperFlex All-NVMe Cluster

This test demonstrates the ability that HyperFlex has leveraged the resources from all nodes in the cluster to satisfy the performance (and capacity) needs of any given VM.

Performance Scaling with Multiple VMs

The seamless pool of capacity and performance presented by the HyperFlex cluster can also be accessed by multiple smaller VMs. In this section, multiple VMs are used that are running a SQL instance with HammerDB.  The cluster setup used is the same as detailed in Table 3.  The details of the VM configuration and workload are given in Table 5  . OLTP workload with 70:30 read write ratio was exerted on each guest VM. The workload stressed each VM with up to 35-40% of guest CPU utilization.

Table 5    VM and Workload Details Used for Multi VM Test

Figure 29 shows performance scaling seen by scaling the number of VMs in the cluster from 4 (1 VM per node), 8 (2 VMs per node) and finally 12 (3 VMs per node). The performance data shown in the following graphs is captured HyperFlex Connect. Note that these are sustained level of performance. Also, deduplication and compression are turned on by default and the default setting is being used in this test as well.  If one or more of these VMs needed additional IOPS / throughout – there will be an increase in storage performance provided the VM itself is not CPU or memory bottlenecked and there is additional performance headroom available in the cluster.

Figure 29     Performance Scaling with VM Count in the Cluster with 4, 8, 12, and 16 VMs

This test demonstrates HyperFlex Data Platform’s ability to scale the cluster performance as we add more SQL virtual machines to the cluster.

All-NVMe Cluster for High Performance Demanding SQL Workloads

This section demonstrates the HyperFlex All-NVMe cluster performance delivered when high storage performance demanding SQL workloads deployed on the cluster. For this test, four large SQL virtual machines (one per node) are stressed to drive higher IOPS. The cluster setup used is the same as detailed in Table 3  . The details of the VM configuration and workload are list in Table 6  . OLTP workload with 70:30 read write ratio was exerted on each guest VM. The workload stressed each VM with up to 55-60% of guest CPU utilization which resulted in 60% of host CPU utilization.

Table 6    VM and Workload Details used for Multiple Large VM Test

Figure 30 shows the performance driven by four large SQL virtual machines (one per node) driving high IOPS from each ESXi node. The test was executed for three hours.

Figure 30     Four Large SQL Server Virtual Machines Test on HyperFlex All-NVMe Cluster

The test described above, demonstrates low latency and consistent performance delivered by a four node HyperFlex All-NVMe cluster as demanded by large SQL virtual machines for a specific database workload simulated by HammerDB tool. You can add additional converged nodes (up to sixteen) to scale IOPS while maintaining IO latencies within reasonable limits.

This section describes the common database maintenance activities and provides a few guidelines for planning database maintenance activities on the SQL VMs deployed on the All-NVMe HyperFlex system.

The most common database maintenance activities include export, import, index rebuild, backup, restore and running database consistency checks on regular intervals. The IO pattern of these activities usually differs from business operational workloads hosted on the other VMs in the same cluster. The maintenance activities would typically generate sequential IO when compared to the business transactions, which generate random IO (in case of transactional workloads). When sequential IO pattern is introduced to the system alongside with random IO pattern, there is a possibility of impact on IO sensitive database applications. Hence caution must be exercised while sizing the environment or controlling the impact by running DB maintenance activities during the business hours in production environments. The following list provides some of the guidelines to run the management activities to avoid the impact on business operations:

·       As a best practice, all the management activities such as export, import, backup, restore and DB consistency checks must be scheduled to run off business hours when no critical business transactions are running on the underlying HyperFlex system to avoid impact on the ongoing business operations. Another way of limiting the impact is to size the system with appropriate headroom.

·       In case of any urgency to run the management activities during business hours, administrators should know the IO limits of hyperconverged systems and plan to run accordingly.

·       For clusters running at peak load or near saturation—when exporting a large volume of data from SQL database hosted on any hyperconverged system to any flat files, it should be ensured that the destination files are located outside of the HyperFlex cluster. This will avoid the impact on the other guest VMs running on the same cluster. For small data exports, the destination files can be on the same cluster.

·       Most of the import data operations will be followed by recreation of index and statistics in order to update the database metadata pages. Usually Index recreation would cause lot of sequential read and writes hence it is recommended to schedule import data in off business hours.

·       Database restore, backup, rebuilding indexes and running database consistency checks typically generate huge sequential IO. Therefore, these activities must be scheduled to run in the out of business hours.

Using a complete guest or database backups, it is not recommended to keep the backups in the same cluster as it would not protect against the scenario where the entire cluster is lost, for example, during a geographic failure, large scale power outage, and so on. Data protection of the virtualized applications that are deployed on the hyperconverged systems are becoming one of the major challenges to the customers. Hence there is a need for most flexible, efficient and scalable data protection platform.

Cisco HyperFlex has integration with several backup solutions, for example, Cisco HyperFlex™ System’s solution together with Veeam Availability Suite gives customers a flexible, agile, and scalable infrastructure that is protected and easy to deploy. More details on Veeam data protection platform is available here: https://www.cisco.com/c/en/us/td/docs/unified_computing/ucs/UCS_CVDs/hyperflexedge_veeam.html

Workloads are rarely static in their performance needs. They tend to either grow or shrink over time. One of the key advantages of the Cisco HyperFlex architecture is the seamless scalability of the cluster. In scenarios where the existing workload needs to grow – Cisco HyperFlex can handle the scenario by growing the existing cluster’s compute, storage capacity or storage performance capabilities depending on the resource requirement. This gives administrators enough flexibility to right size their environment based on today’s needs without worrying about future growth.

This section contains the following:

·       High SQL Guest CPU Utilization

·       High Disk Latency on SQL Guest

VMware lists the common troubleshooting scenarios for the virtual machines hosted on VMware ESXi cluster here: https://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=2001003

This section discusses some of the commonly seen performance problems and the methodology followed to solve those. 

Some of the commonly seen performance problems on virtualized/ hyperconverged systems are detailed below.

High SQL Guest CPU Utilization

When high guest CPU utilization with lower disk latencies on SQL guest VMs is observed and CPU utilization on ESXi hosts appears to be normal, then it might be the case that virtual machine is experiencing a CPU contention. The solution to overcome this is to add more vCPUs to the virtual machine as the workload is demanding more CPU resources.

When high CPU utilization is observed on both guest and Hosts, then one of the options to be looked at is upgrading to a higher performing processor. More options to solve this issue is mentioned in VMware vSphere documentation here: https://pubs.vmware.com/vsphere-60/index.jsp#com.vmware.vsphere.monitoring.doc/GUID-5F8147A1-6416-4D29-BA3D-E4CED3966016.html.

High Disk Latency on SQL Guest

The following guidelines can be used to troubleshoot when higher disk latencies are observed on SQL guest VMs:

·       Use ESXTOP charts to identify the guest latencies versus kernel latencies and follow the options mentioned in section Deployment Planning.

·       If a higher HX storage capacity utilization nearing expected thresholds occurs (above 60% usage), SQL VMs might also experience IO latencies at both guest and kernel levels; it is recommended to scale up the cluster by adding a new HX node to the cluster.

The Cisco HyperFlex HX Data Platform revolutionizes data storage for hyperconverged infrastructure deployments that support new IT consumption models. The platform’s architecture and software-defined storage approach gives you a purpose-built high-performance distributed file system with a wide array of enterprise-class data management services. With innovations that redefine distributed storage technology, the data platform provides you the optimal hyperconverged infrastructure to deliver adaptive IT infrastructure. Cisco HyperFlex systems lower both operating expenses (OpEx) and capital expenditures (CapEx) by allowing you to scale as you grow. They also simplify the convergence of compute, storage, and network resources.

HyperFlex All-NVMe storage system is a purpose build and co-engineered system to cater various enterprise grade applications such as databases which demands low latency and consistent performance. The performance lab tests showed All-NVMe cluster capabilities for delivering low latency and consistent performance for Microsoft SQL Server OLTP database workloads.

Lab solution resiliency tests detailed in this document show the robustness of the solution to host IO sensitive applications like Microsoft SQL Server. The system performance tuning guidelines described in this document addresses the platform specific tunings that will be beneficial for attaining the optimal performance for a SQL server virtual machine on Cisco HyperFlex All-NVMe System. The performance observed on Cisco HyperFlex systems focused in this document proves Cisco HyperFlex All-NVMe system is an ideal platform to host high performing low latency applications like Microsoft SQL Server database.

Gopu Narasimha Reddy, Cisco Systems, Inc.

Gopu is Technical Marketing Engineer with Cisco UCS Data Center Solutions. He has over 13 years of experience focusing on customer driven solutions for Microsoft SQL Server databases on various Operating systems and Hypervisors. His areas of interest include building and validating reference architectures, publishing TPC benchmarks, development of sizing tools in addition to assisting customers in SQL deployments.

Acknowledgements

For their support and contribution to the design, validation, and creation of this Cisco Validated Design, the author would like to thank:

·       Babu Mahadevan Venkata Subramanian, Cisco Systems Inc.

·       Jeffrey Buell, Cisco Systems Inc.

·       Mithlesh Thukral, Cisco Systems Inc.