Cloud Service Assurance for Virtualized Multiservice Data Center Design Guide
CLSA VMDC System Architecture
Downloads: This chapterpdf (PDF - 4.59MB) The complete bookPDF (PDF - 4.31MB) | Feedback

CLSA VMDC System Architecture

Table Of Contents

CLSA VMDC System Architecture

Functional View

Component View

System Components

Monitored Components and Services

Key Functions

Automatic Enablement of Service Assurance

Automatic Discovery

Zenoss APIs for Programmatic Provisioning

Fault Performance, Configuration Data Collection, and Device Modeling

Event Processing

Root Cause Analysis and Service Impact Analysis

Zenoss SIA and RCA

VMDC Assurance Service Models

VMDC RCA and SIA Use Cases

Northbound Interface

SNMP Northbound Interface

Zenoss SNMP Notification Content

Zenoss Notification Filtering

Zenoss Service Impact SNMP Trap

WS or ReST API

Northbound Integration Use Case Examples

Performance Management

Dashboards

Reporting

Multiservices


CLSA VMDC System Architecture


This chapter provides an overview of the Cloud Service Assurance for VMDC (CLSA VMDC) system architecture.

Functional View and Component View provide the functional and component views of the CLSA VMDC system architecture.

System Components defines the components and interfaces used to deliver the system functions.

Monitored Components and Services lists the VMDC devices that are monitored by CLSA VMDC.

Key Functions defines the functions of the new architecture.

Functional View

Figure 3-1 illustrates the functional framework for CLSA VMDC. This functionality is delivered with one or more of the integrated products/components. In CLSA VMDC, only a subset of this functionality is available. This section defines the functional layers of this architecture and identifies the layers that are available in CLSA VMDC.

Figure 3-1 Functional View of CLSA VMDC Architecture

The Managed Device Layer consists of Data Center (DC) infrastructure including compute, storage, and network components with instrumentation for inventory, fault, and performance data collection. The instrumentation used in this system includes Simple Network Management Protocol (SNMP), syslog, XML Application Programming Interface (API), NETCONF, vSphere API, and so on. Details of interfaces used per VMDC component are included in Monitored Components and Services.

The Domain/Element Management Layer includes the UCS Manager (UCSM) and vCenter. They provide intra-domain inventory, fault, and performance monitoring for UCS and VMware hosts and VMs. These domain managers offer northbound interfaces APIs as well as SNMP and syslog interfaces. CLSA VMDC utilizes UCS XML API and vSphere API interfaces. CLSA VMDC 3.0 also introduces the Storage Management Initiative Specification (SMI-S) Provider domain manager to incorporate EMC VMAX and VNX inventory, fault, and performance monitoring.

The Service Assurance Manager (SAM) Layer provides all inter-domain functions and a single pane of glass to monitor all VMDC domains including compute, storage, and network. The high-level functions of each of the SAM layers are as follows:

Data Collection Layer. This layer leverages domain managers, third-party tools, and so on to obtain performance, availability, and event data for the end-to-end multi-domain system via a range of open protocols such as SNMP, SSL, WMI, and so on. This layer is responsible for normalizing this data into a consistent format and persisting data. Collected data includes inventory, fault, and performance type of information.

Modeling Layer. This layer performs discovery, classification, and modeling to determine component dependencies and service dependency graphs. Both performance and fault data should be included in device and service models.

Service Model-based Technology. CLSA VMDC uses service model-based technology which is described in more detail in Root Cause Analysis and Service Impact Analysis and Zenoss Cloud Service Assurance Overview, page 4-1CLSA VMDC.

Root Cause Analysis (RCA). Leverages the dependency graph or analytics algorithms to determine which events are the probable root cause of the problem and which ones are just consequences that create noise. Therefore, RCA reduces Mean Time to Repair (MTTR). There are a number of different approaches to RCA, but most of them can be classified in one of the following technologies:

1. Event correlation rules-based

2. Topology and service model-based

3. Analytics based

Service-Impact Analysis (SIA). Leverages the dependency graph or analytics algorithms and collects fault and performance data to do the following:

Answer who is impacted by the failures

Prioritize urgency of failure tickets based on business relevance

Determine whether redundancy protected the service

Identify failure impacted customers/tenants

Prevent future failures by identifying potential service impacting technical risks before they impact service

Provide data for SLA measurements and reporting

Performance Aggregation Layer. This layer aggregates performance data from multiple domains (e.g, storage, network, compute for VMDC), normalizes it in the same format and units, provides threshold crossing alerts to the fault management part of the SAM, trends the data over time, and in some cases, performs additional analysis of the data.

Presentation Layer. This layer provides a single view to do both fault and performance monitoring for the entire system. Presentation is done both via dashboards and reports. CLSA VMDC includes SP dashboards for both fault and performance.

Northbound Interface. The Northbound Interface (NBI) is a special form of the presentation layer where normalized and enriched data is presented to northbound OSS/BSS systems via open interfaces such as WS API, SNMP, and email.

Component View

This section defines the components used to deliver those functions, as well as their interfaces. The key component of the architecture for CLSA VMDC is Zenoss Cloud Service Assurance (CSA), which plays the role of the SAM. In addition, several domain managers are utilized - UCS Manager (UCSM) for UCS hardware monitoring, VMware vCenter for monitoring the virtualized infrastructure, and SMI-S Provider for EMC VMAX and VNX monitoring.

Figure 3-2 illustrates the components and interfaces used to deliver the functional layers of the CLSA VMDC architecture.

Figure 3-2 Component View of CLSA VMDC Architecture

Key system interfaces include:

Southbound interface instrumentation to collect data from managed system devices.

Northbound interface to integrate with OSS/BSS systems such Manager-of-Managers (MoM) (e.g., IBM Netcool), ticketing systems (e.g., Remedy) and so on. The interfaces available from CLSA VMDC are SNMP, JSON API, email, page, commands, and Advanced Message Queuing Protocol (AMQP).

CLSA VMDC offers the JSON API interface for integration with orchestration and fulfillment systems.

System Components

Table 3-1 lists the Cisco and third-party components used in CLSA VMDC.

Table 3-1 Cisco and Third-Party Components Used in CLSA VMDC 

Vendor
Model
Description

Zenoss

Resource Manager 4.2.3

Zenoss CSA software module that performs resource discovery, monitoring, and modeling.

Zenoss

Impact 4.2.3

Zenoss CSA software module that performs service impact discovery and analysis.

Zenoss

Analytics 4.2.3

Zenoss CSA software module that performs long term data trending, processing, and reporting.

vCenter

vCenter 5.0

Domain manager for VMware based virtualization

Cisco

UCSM 2.0

Domain manager for UCS platform

EMC

SMI-S Provider 4.5.0.1

Domain manager for EMC VMAX and VNX platforms



Note The Zenoss software modules are packaged together as Zenoss CSA 4.2.3.


Monitored Components and Services

Table 3-2 lists the VMDC devices that are monitored by the CLSA VMDC system out-of-the-box and the instrumentation (interfaces) utilized by Zenoss CSA to collect data.

Table 3-2 VMDC Components Monitored by CLSA VMDC 

Managed Component
Interfaces Utilized in CLSA VMDC 3.0

Compute Components

UCS 5108; B-series blades

ICMP, UCSM XML API

UCS 6100, 6200

ICMP, UCSM XML API

VMware ESX and ESXi Hypervisors

ICMP, vSphere API

VMware Virtual Machines

ICMP, vSphere API

Storage Components

MDS 9000

ICMP, SNMP

EMC VMAX 1

ICMP, SMI-S API

EMC VNX 1

ICMP, SMI-S API

FAS6080, FAS3000

ICMP, SNMP, SSH

Network Components

UCS 6100, 6200

ICMP, UCSM XML API

Nexus 7000 (e.g., 7018, 7010, 7009, 7004 including M1 and F1/F2 cards) 3,4

ICMP, NETCONF, SNMP

Nexus 5000 (e.g., 5548, 5596, and 5020)

ICMP, NETCONF, SNMP

Nexus 3000 1

ICMP, NETCONF, SNMP

Nexus 2000 (e.g., 2248 and 2232)

ICMP, NETCONF, SNMP

Nexus 1000V / Nexus 1010 1

ICMP, NETCONF, SNMP

ASR 9000

ICMP, SNMP, SSH

ASR 1000

ICMP, SNMP

Network Services Components

Catalyst 6500 VSS

ICMP, SNMP, SSH

ACE (e.g., ACE20, ACE30, ACE4710 2)

ICMP, SNMP, ACE XML API

FWSM

ICMP, SNMP

ASASM 1

ICMP, SNMP

ASA 5555 2

ICMP, SNMP

ASA 5580-40

ICMP, SNMP

ASA 5585-40

ICMP, SNMP

Virtual Security Gateway

ICMP, SNMP, NETCONF, SSH


1 Denotes new enhancement for CLSA VMDC 3.0.
2 Denotes new enhancement for CLSA VMDC 2.3.
3 FabricPath F1 cards are added to Nexus 7000 devices for CLSA VMDC 3.0.
4 FabricPath F2 cards are added to Nexus 7000 devices for CLSA VMDC 2.3.

For detailed information on software releases, please refer to the following documents:

VMDC 2.2 Implementation Guide

VMDC 2.3 Implementation Guide

VMDC 3.0 Implementation Guide


Note Information on previous VMDC system releases can be found at VMDC System Releases.


Key Functions

This section describes the key functions of CLSA VMDC.

In the overall lifecycle of assurance services, the first task that has to be completed is enablement of service assurance services. Automatic Enablement of Service Assurance provides details about enabling service assurance, including provisioning and automatic discovery. Once assurance services are enabled, they can be used for Day 2 operations. Figure 3-3 illustrates and explains the high-level, end-to-end data flow through the fault and problem management part of CLSA VMDC.

Figure 3-3 End-to-End Fault and Problem Management Data and Processing Flow

The following sections discuss each of the stages and functions in this sample data flow:

Fault Performance, Configuration Data Collection, and Device Modeling

Event Processing

Root Cause Analysis and Service Impact Analysis

Northbound Interface

This section also discusses the following additional functions related to the overall platform and its use:

Performance Management

Dashboards

Reporting

Multiservices

Automatic Enablement of Service Assurance

Automatic enablement of service assurance can be achieved in a couple of different ways. Fundamentally, the following are approaches that can be taken to automate service enablement and life cycle:

1. Reduce necessary amount of configuration (by using technology that is self learning (e.g., self learning thresholds))

2. Automatic discovery (by assurance system)

3. Programmatic orchestrated provisioning (via integration with orchestration system)

CLSA VMDC focuses on automatic discovery. CLSA VMDC also provide APIs for programmatic orchestrated provisioning, but they are not integrated or validated with any particular orchestration system. Automatic discovery and APIs are discussed in the following sections.

Automatic Discovery

Zenoss APIs for Programmatic Provisioning

Automatic Discovery

The following types of objects are automatically discovered in CLSA VMDC:

Monitored devices (e.g., UCS, Nexus 7000, MDS 9000, etc.)

Sub-components of devices and their relationships (e.g., UCS chassis, blades, fabric interconnect, etc.)

Tenant-based Service Impact Analysis (SIA) model for the compute (e.g., tenant Virtual Machine (VM) mapping to service impacting resources, both dedicated and shared vCenter and UCSM managed resources). The exception is tenant name and its link to the service, which cannot be discovered, but relies on orchestrated provisioning. In this release, tenant name and mapping to the VM are provisioned manually, but the API is provided.

Figure 3-4 and Figure 3-5 illustrate examples of automatic enablement of service assurance.

Figure 3-4 Real-time Automatic Discovery of Device Components - Cisco UCS

Figure 3-5 Real-time Automatic Discovery of Tenant Service Impact Model

Zenoss APIs for Programmatic Provisioning

CLSA VMDC offers APIs to programmatically provision the following components in the service impact tree:

Tenant Name

Tenant ID

Service Name

Service ID

VM Name

VM ID

This enables automatic onboarding of the tenant and tenant compute service, which maps them to the already automatically discovered VM and its relationships to shared hardware.


Note Proof of Concept (PoC) of this functionality integrated with the Cisco Intelligent Automation for Cloud (CIAC) orchestration stack has been performed by Cisco Advanced Services; however, it was not validated as part of the CLSA VMDC system. If this functionality is desired in the field before it is included as part of the Systems Development Unit (SDU) system release, then Cisco Advanced Services can perform integration with the desired orchestration stack using the provided API.


Figure 3-6 illustrates the high-level workflow that provisions the tenant and tenant service and then maps the workflow to the automatically discovered VM and the rest of the automatically discovered infrastructure.

Figure 3-6 Zenoss Tenant Provisioning Using CIAC Orchestration

Fault Performance, Configuration Data Collection, and Device Modeling

Consolidated monitoring and data collection at the SAM layer is one of the fundamental functions of CLSA VMDC. Consolidated monitoring and data collection is characterized by the following attributes:

Various domains (applications, compute, storage, network). The cloud assurance system needs to provide a single pane of glass to monitor components from various domains.

Fault and performance data. The cloud assurance system needs to consolidate fault and performance data and leverage both for all of its higher order functions like RCA and SIA.

Various data sources, interfaces, and protocols. The cloud assurance system needs to collect data from multiple data sources and protocols and consolidate this data in unified device and service models. Some examples of different data sources and protocols are SNMP, syslog, WS API, Netflow, customer opened tickets, and so on.

Zenoss Data Collection

Zenoss CSA offers consolidated monitoring for VMDC, including consolidation of domains (i.e., support for OS, compute, storage, and network), consolidation of performance and fault data (i.e., takes into consideration both polled performance data, asynchronous events it receives, as well as synthetic events it generates for both performance and availability), and consolidation of data sources (i.e., device monitoring models utilize multiple data sources such as SNMP, syslog, API, and consolidate it within unified device model).

Zenoss CSA uses an agentless data collection approach, which is critical for the type of scale expected in cloud systems. Instead of installing an agent on monitored devices, Zenoss supports a rich set of protocols to enable data collection. A list of protocols used for data collection from VMDC devices is included in Monitored Components and Services. The following is a more comprehensive list of data collection interfaces that the Zenoss CSA platform supports:

Event input:

SNMP

Syslog

XML Remote Procedure Call (RPC)

JavaScript Object Notation (JSON)/API

AMQP

Windows Event Log

Easily configurable protocol usage:

Secure Shell (SSH)

Java Management Extensions (JMX)

Windows Management Instrumentation (WMI)

Perfmon

Any script that returns data in a known format (such as Nagios)

Other collection mechanisms (model/performance/event data):

Internet Control Message Protocol (ICMP)

Telnet

JMX

Hypertext Transfer Protocol (HTTP) - Web Transactions

Oracle

Structured Query Language (SQL) Server

MySQL

Apache (mod_status)

memcache

Splunk Queries

Simple Mail Transfer Protocol (SMTP)

SMI-S Provider

Post Office Protocol (POP)

UCSM XML API

vSphere Simple Object Access Protocol (SOAP) API

vCloud Director

Amazon EC2 and CloudWatch

Cisco CallManager (AXL)

Domain Name System (DNS)

Lightweight Directory Access Protocol (LDAP)

Network Time Protocol (NTP)

File Transfer Protocol (FTP)

Internet Relay Chat (IRC)

Extensible Messaging and Presence Protocol (XMPP)

Remote Procedure Call (RPC)

Network News Transfer Protocol (NNTP)

Zenoss Device Modeling

Device modeling in Zenoss goes beyond traditional device discovery; it also uses standard Management Information Bases (MIBs) to discover interesting aspects of the device and automatically defines models for that device type. Once modeled, these learned attributes can be inherited as part of the model when a new device of the same type is discovered again. The information below describes various attributes of the Zenoss device modeling process.

Initial Zenoss Model (plugins):

Interfaces to access device and objects of interest (KPI statistics, events, thresholds, etc.) are statically defined

Models are assigned to a device class

Device Modeling:

During individual device discovery, all modeler plug-ins for the device class are automatically considered, and a model per instance of the device is created.

After discovery modeling, monitoring and event processing automatically starts.

Device Remodeling:

Model per device instance can dynamically change in response to events (e.g., blade removed, etc.)

ZenModelerDeamon - per collector configuration happens every 12 hours

ZenVMwareDeamon (exception for VMware and remodels every 4 hours)

List of events that trigger remodeling is configurable (default set exists)

An example of unified monitoring using Zenoss CSA is illustrated in Figure 3-7.

Figure 3-7 Unified Monitoring Using Zenoss CSA

Event Processing

In CLSA VMDC, event processing is divided into two categories:

Basic event processing

Event processing that is part of RCA and SIA

This section only describes basic event processing functions, while RCA and SIA are discussed in the following sections. The basic event processing functions included in this system are event classification, normalization, de-duplication, enrichment, persistence, and clearing.

Event classification groups similar events in event classes, so that some of the more complex processing may be simplified by looking at event classes rather than each individual event.

Event normalization translates various formats of the raw collected data into a single format that is used by the SAM. Often, the same format or subset of the fields of normalized format can be sent to northbound systems. This function allows simplified integration of northbound systems since they have to deal with a single event format for multiple device types and instrumentation protocols.

Event de-duplication eliminates multiple events that have the exact same content with the exception of the time stamp. After de-duplication, a single event is kept, and typically a counter indicating the number of occurrences of the event is added, as well as a timestamp indicating the first and last occurrence of the duplicate event.

Event persistence archives all events to be used for forensic analysis. In some systems, persistence exists only on post-processed events, while in others, for raw events as well.

Event clearing is used to indicate when the original condition for which the event was raised is removed. Explicit event clearing is done by generating clearing events with the field within the clearing event, which points to the ID of the event that it is clearing. For example, if an interface down event for a specific interface had an ID of ID1, when the interface goes up again, an event with ID2 should be raised, which includes as one of its fields a reference to event ID1. Explicit event clearing is recommended. In addition to explicit clearing, time-based clearing can be utilized as well. Time-based clearing clears the event after a specific time interval elapses from the time that the original event was received.

Root Cause Analysis and Service Impact Analysis

One of the key functions of CLSA VMDC is Root Cause Analysis (RCA) and tenant-based Service Impact Analysis (SIA).

The objective of RCA is to reduce MTTR by determining which events are probable root causes of the problem and which events are just consequences that create noise.

The following are the objectives of tenant-based SIA:

To prioritize the urgency of failure tickets based on business relevance.

To determine whether redundancy protected the service.

To identify failure impacted customers/tenants.

To prevent future failures by identifying potential service impacting technical risks before they impact service.

To enable Service Level Agreement (SLA) measurements and reporting.

The following sections are detailed:

Zenoss SIA and RCA

VMDC Assurance Service Models

VMDC RCA and SIA Use Cases

Zenoss SIA and RCA

Zenoss CSA uses model-based SIA, which produces a set of ranked probable root causes as a by-product of SIA. This service impact-based approach to RCA is a fundamentally different approach from legacy rule-based systems:

Bottom-up. What services are impacted by conditions below (Zenoss) vs.

Top-down. What is the cause of problem at service level (legacy products)

Zenoss does not determine a single root cause, but instead identifies multiple related events (probable root cause events) and presents the following:

A root cause ranking algorithm is utilized to rank probable root cause events in order of confidence that the event is the actual root cause event. This algorithm ranks impact events based on a variety of criteria, including the severity of the event, service graph depth, and the number of graph branches affected by an event.

Hierarchical service dependency graphs provide a visual indication of probable root causes leading to a service impact.

Events flow through the graph referencing molecular node policies to determine whether they should be passed, filtered, aggregated, or masked. There are a few key elements of RCA and SIA in Zenoss CSA. Each assurance service within Zenoss is modeled with a service impact tree that consists of a set of nodes, policies applied to the nodes, and the relationships between the nodes:

The service can be arbitrarily defined and can be a very abstract service that consists of other sub-services, or on other extreme, one can even define a single physical interface as a service. This provides a very flexible framework for service definition.

Model nodes represent arbitrary components such as physical, logical, or virtual resource. For example, nodes can represent an end-to-end service such as voice, a virtual resource such as a VM, or a physical resource such as a chassis or physical interface. The following four types of nodes are currently supported, as illustrated in Figure 3-8:

Physical. Systems, infrastructure, and network devices that a service relies on.

Virtual. Software components that make up a service.

Logical. Aspects of a service that must be measured or evaluated as a set to determine state (facilitates extension of an impact graph by providing a hook to incorporate arbitrary events into impact analysis).

Reference (future release). Provide a link to dependencies managed by an external instance of Zenoss or other management system capable of propagating state information to Zenoss.

Figure 3-8 Node Types

Policy is defined per node, which allows it to move as the resources move, which is a critical characteristic for the cloud environment. Zenoss refers to this policy as a molecular policy since it is defined per node. Zenoss utilizes a very simple policy that can define the state of the node solely as a function of the state of its children nodes, which allows for service impact "rules" decoupling from device events resulting in the following:

"Rules" defined in a single place for any given device or service: device events processing in event processing software modules, service impact processing in service impact graphs (i.e., device events do not need to be considered in service level rules)

Simplified development and maintenance of cross-domain service impact and RCA customizations: do not have to correlate device events from multiple devices to determine cross-domain service impact and possible root causes

Note that whenever desired, device events can be used as part of service impact "rules" via use of logical nodes whose rules define how to interpret the service impact of specific events based on its type and severity.

Policy can be global or contextual:

Global policy applies to device/service type in any service graph.

Contextual policy applies only to device/service in the particular service graph.

Each node has a default policy applied, which reduces the need for custom configuration. The default policy is often sufficient, but can be modified where required via GUI or API. Figure 3-8 illustrates a sample node policy.

VMDC Assurance Service Models

In order to perform SIA, CLSA VMDC uses service models with polled and asynchronous data to perform SIA and RCA. CLSA VMDC offers an out-of-the-box tenant service model for compute. In future releases, CLSA VMDC will expand the library of out-of-the-box service models that will be validated and delivered as part of this system, however, note that users can easily customize service models as well as create new ones.

Tenant Compute Assurance Service

Figure 3-9 defines the out-of-the-box tenant compute service model to be delivered as part of CLSA VMDC. More details are provided about this service model in Zenoss SIA and RCA.

Figure 3-9 Tenant Compute Assurance Service Model - Generic Application

Service Model Policy

Each node (referred to as the parent node) in the service model has a policy defined that calculates the state of that node based on the state of its children and any explicit events associated with the parent node.

For the particular service model illustrated in Figure 3-9, the specific policies listed in Table 3-3 should be applied.

Table 3-3 Service Model Policy Decisions 

Node
Node State
If Child Node State

Tenant Compute Service

UP/DOWN/AT RISK

UP/DOWN/AT RISK

Tenant Guest OS

UP/DOWN/AT RISK

UP/DOWN/AT RISK

Tenant VM

UP/DOWN/AT RISK

UP/DOWN/AT RISK

ESXi Cluster

UP/DOWN
AT RISK

All Children up/down
At least One Child Down/At Risk

ESXi Host

UP/DOWN/AT RISK

UP/DOWN/AT RISK

UCS Blade

UP/DOWN/AT RISK

UP/DOWN/AT RISK


Out-of-the-box, all nodes use the default policy where the worst impact wins. The one exception is the VMware cluster, which is DOWN if all children are DOWN and DEGRADED if any nodes are DOWN or DEGRADED.

In addition to considering the parent/child policy, the explicit state of the nodes is determined by both availability and events for components the node represents. For VMware and UCS nodes, the explicit node impact status is determined mainly by modeled properties. As modeling occurs or various events are received, Zenoss reassesses the impact state by querying the Zenoss model. For example, when a VM power off event is received, the model is updated and the VM status is reassessed and updated.

Service Model Variations

Note that the model defined in this section illustrates a single-tier application with a single VM. Variation of this service model would be models for the following:

Multi-tier application, where there would be multiple "tenant dedicated VM" blocks tied to the tenant compute service. The tenant compute service default policy may need to be customized.

Single-tier application that supports application level redundancy via clustering (e.g., Cisco UC applications such as CUCM). In this case, the model would be modified to include multiples of "tenant dedicated VM" blocks. The default policy used for the "tenant compute service" should be applicable. An example of this service model is illustrated in Figure 3-9.

Service Model Enablement

Most of this model is automatically discovered, while the top node of the service model needs to be provisioned. Typically, provisioning would be done in an automated way when the tenant and VM get onboarded. In CLSA VMDC, there is no integration with the orchestration stack, and as such, the top node of the service model is manually provisioned. Note that in real deployments, per-tenant manual provisioning is not an option, in which case either an available Zenoss API can be used by the orchestration platform of choice, or if not provisioned, the tenant service impact is still possible but results are given in the VM context rather than tenant service context. For example, there would be no automatic mapping between tenant name, tenant service name, and VM ID.

In future CLSA VMDC releases, integration with VMDC orchestration stacks will be implemented and validated. In addition to automatic discovery of the service model from VM down, if operating systems such as Windows or Linux are deployed, they should also be automatically discovered.

Mobility Handling

The host to VM relationship is given by VMware during modeling stage. Whenever VMware generates an event that indicates VM movement, Zenoss reacts and remodels the source and target hosts to update its model. Depending on the event collection interval specified in the Zenoss configuration, the model change can take anywhere from 15 seconds to 3 minutes. With the out-of-the-box configuration, the average time would be about 1 minute.

Redundancy Implications

A service model with three service states accounts for redundancy. The AT RISK state is used to indicate conditions where the service or service model node is still functioning despite a failure of one of its children because redundancy protected the service. For the particular service model shown in Figure 3-9, redundancy capabilities that are accounted for include the following:

If one of the blades/hosts fails, and the vCenter cluster that VM belongs to has multiple blades/hosts, then the VM node is marked AT RISK as opposed to DOWN based on the status of its children. Note that explicit VM related state and events can result in the state of the VM node being down even though the state of its children alone would result in an AT RISK state

In a case where there is application level redundancy and thus more than one VM and application deployed for single tier applications, there is also service model redundancy built in on the application/VM level. For example, a service is AT RISK if one of the application nodes/VMs is DOWN because the remaining application/VM nodes provides redundancy for the failed application/VM node.

VMDC RCA and SIA Use Cases

Once service impact models are defined, the data is applied to service impact models to maintain real-time state of the service availability and performance, as well as to determine probable root cause of any failures that may happen. This section provides a list of failure scenarios (use cases) validated as part of the CLSA VMDC test effort, for which the out-of-the-box compute service model can determine correct probable root cause and service state for previously defined services. All of the use cases are validated in an environment where VMware High Availability (HA) is deployed.

Refer to Root Cause Analysis and Service Impact Analysis for an example workflow illustrating a UCS switch failure event, including screen shots.

Use Case Name (Fault):

VM Failure

VM vNIC failure

VM vMotion - VM vMotion is not a true fault event, since the VM stays up, however, the impact graph does track the VM's host swap.

ESXi host failure

UCS Blade failure

UCS chassis failure

UCS P/S failure

UCS FEX failure

UCS 6100 chassis failure

UCS 6100 interfaces to UCS 5100 failure

VM CPU degradation (Threshold Crossing Alert (TCA))

VM Memory degradation (TCA)

Host CPU degradation (TCA)

Host Memory degradation (TCA)

Northbound Interface

One of the key, new functions of CLSA VMDC architecture is a single, normalized Northbound Interface (NBI) provided by the SAM.

The key objectives of the single, normalized interface are:

To simplify and reduce the cost of integrating providers existing northbound system with the CLSA VMDC system. The provider needs to integrate and maintain just one interface rather than multiple dozens of interfaces towards individual devices and/or domain managers. CLSA VMDC is responsible for absorbing updates related to any relevant changes in the underlying system and devices.

To enable CLSA VMDC to be inserted in various business and operational deployment environments. This is achieved by offering a variety of interface protocols, rich filtering capabilities, and notifications with tenant awareness.

To enable CLSA VMDC to simplify service assurance of overlaid application based systems that are deployed on top of VMDC infrastructure. An example of this type of system is the Hosted Collaboration Solution (HCS). This is achieved by providing tenant service level notifications rather than device level notifications, which enables a service overlay (or multi-tier SIA) to be implemented by HCS, and as such, Cloud Service Assurance-HCS (CLSA-HCS) would have to deal with the state of only a handful of services coming from CLSA VMDC, rather than thousands of events coming from individual VMDC devices.

Zenoss northbound integration is supported via:

JavaScript Object Notation (JSON)/Representational State Transfer Application Programming Interface (ReST API)

SNMP Traps (ZENOSS-MIB.txt and ZENOSS-IMPACT-MIB.txt)

Syslog

Event queues (AMQP and Java/Python wrappers) and event commands (command line call with event context)

SMTP email

Configurable filtering capabilities are offered to provide different data to different northbound consumers. The following sections describe the interfaces, data, and filtering capabilities in more detail.

Sections

SNMP Northbound Interface

Zenoss SNMP Notification Content

Zenoss Notification Filtering

Zenoss Service Impact SNMP Trap

WS or ReST API

Northbound Integration Use Case Examples

SNMP Northbound Interface

One of the key requirements for CLSA VMDC is to offer asynchronous notifications via SNMP. These notifications are consumed either by the provider's existing northbound systems such as MoM, ticketing, and SLA management systems, or by other Cisco systems deployed on VMDC architecture such as HCS.

Regardless of the source or type of the event, all events should be sent using the same normalized format, however, as discussed in this chapter, there may be differences in the values of the populated fields based on the type of events (e.g., service impact events contain information about service name and state, while device level events do not).

Zenoss SNMP Notification Content

Zenoss CSA uses custom Zenoss MIB implementations for northbound notifications. The original SNMP MIB addresses the resource manager part of the product, but not the service impact part. MIB extensions have been designed to address service impact events and related probable root cause events as a part of the this phase of CLSA VMDC. For a discussion of the new service impact trap, see Zenoss Service Impact SNMP Trap.

Events associated with devices use ZENOSS-MIB for notifications. The ZENOSS-MIB.txt file is located in the following Zenoss directory: $ZENHOME/share/mibs/site. Device level SNMP notifications can be sent to multiple destinations. Refer to the Zenoss Cloud Service Assurance Installation and Administration Guide for more information regarding notifications.

Table 3-4 maps the fields of Zenoss MIBs to the SAM requirements.

Table 3-4 Zenoss MIB Fields

Zenoss MIB Field Name
Description

evtId

Unique identifier ID of the event

evtDedupid

De-duplication ID of the event

evtDevice

Device associated with event

evtComponent

Device component associated with event

evtClass

Event classification

evtKey

Event key used for refining event granularity beyond device and component. Used in de-duplication, automatic clearing.

evtSummary

Event message truncated to 128 characters

evtSeverity

Event severity number: 0=clear(normal), 1=debug, 2=info, 3=warning,4=error, 5=critical

evtState

Event state number: 0=new, 1=acknowledged, 2=suppressed

evtClassKey

Class key for rule processing often matches component

evtGroup

Logical grouping of event sources

evtStateChange

Last time event changed through administrative activity

evtFirstTime

First time an event was received

evtLastTime

Last time an event was received

evtCount

Number of times this event has been seen

evtProdState

Production state of the device or component associated with this event

evtAgent

Collector process that received or created this event

evtDeviceClass

Class of device that this event is associated with

evtLocation

Location of device that this event is associated with

evtSystems

Systems containing the device that this event is associated with

evtDeviceGroup

Groups containing the device that this event is associated with

evtIpAddress

IP address that this event was generated or sent from

evtFacility

Syslog facility if the event was initially sent as a syslog

evtPriority

Syslog priority if the event was initially sent as a syslog

evtNtEvId

Windows NT_EVENT_ID if the event was initially received from Windows event log

evtOwnerId

User that acknowledged this event

evtClearId

evtId that cleared this event

evtDevicePriority

Priority of the device that this event is associated with

evtClassMapping

Name of the event class mapping that matched this event


Zenoss Notification Filtering

Filtering capabilities using Zenoss Triggers can be used to customize notifications based on the needs of different northbound consumers:

Multiple subscribers/receivers may receive notifications.

Each notification subscriber/receiver may apply a different filter: one receiver may subscribe to service events, another may subscribe to compute events, and a third may subscribe to network events.

Each system user should be able to apply different filters.

For more information regarding Triggers, refer to the Zenoss Cloud Service Assurance Installation and Administration Guide.

Zenoss Service Impact SNMP Trap

This section defines the SNMP notification for Zenoss Impact, which is new for CLSA VMDC 3.0. The following data is available internally within Zenoss Impact for service related events. This data was used by the notification script in CLSA VMDC 2.2.

Service Name

Severity

Timestamp

Service state

URLs to EventDetail, page to acknowledge and close events, device events

All events in the impact chain. Each event in impact chain includes:

Device

Component

Device Class

Event Class

Severity

Timestamp

Message

URLs to EventDetail, page to acknowledge and close events, device events

Zenoss Impact provides a flexible framework to define arbitrary services, including support for hierarchical service nesting. In such environments, the question arises for which nodes and/or levels of hierarchy notifications should be sent. Services are collected under Service Organizers. A Service Organizer consists of multiple folders and in each folder there is set of services. In Zenoss Impact, the notification trigger criteria is configured for Service Organizer folders and its services and not based on individual nodes and their hierarchy level in the impact tree. This approach provides good balance between flexibility to select notification trigger criteria and simplicity of implementation.

For CLSA VMDC to send notifications per service instance state change, the appropriate structure must be created to organize the services. The following sections discuss the folders and the structure used for the services defined in CLSA VMDC 3.0.

Service Organizers

Service Organizers are located on the left tab in the Impact GUI.

The Shared Services folder includes:

Service Name X (e.g., Network Aggregation service, Network Core service, etc.)

Service Name Y

The Customer Name folder includes:

Tenant Service 1 (e.g., Tenant Compute service, Tenant Network service, etc.)

Tenant Service 2

Notification Triggers

The user is able to select services and or/folders for which to send notifications. This action is available both in the GUI, as well as via the REST API so that the orchestration system at the time of onboarding the tenant service can select whether or not to enable notification for the service.

The notification policy should be selectable both per folder or per service instance. This enables support for the following use cases:

Where a single operator or NB system manages and/or provides visibility to all services of single tenant/customer (since one folder is defined per tenant).

Where different services of the same tenant are managed by different operators/NB systems, e.g., notification for IaaS services are sent to the IaaS operator while notifications for Unified Communications as a Service (UCaaS) services are sent to the UC operator.

Notification Timing

This section defines the guidelines for service impact notification triggers and timing. An attempt is made to balance any delay in notifications indicating change with excessive noise in events sent due to transient state during service impact analysis. In order to have the capability to delay some service impact notifications, there is a timer that can be configured (value range 0-10 minutes with default of three minutes).

Service impact notifications are triggered when the following events occur:

If the service state changes (top-level service in each folder):

The notification indicating a service state change should always be sent immediately, regardless of the value of the notification delay timer. This enables the northbound system to immediately detect the change. Also, for northbound systems that are using service state notifications to measure service availability and SLA, this immediate notification enables more accurate service availability measurements.

When the service state changes back to UP, the event should serve as a clearing event for the previous service state change event. As such, the ID of the service event that it is clearing must be included.

If the service state does not change, but most a probable root-cause event changes (i.e., root cause ranked with highest confidence% changes):

This trigger honors the notification delay timer, and as such, it is sent only if the event is generated after the notification timer expires.

The following example shows the use of the notification delay timer and the two notification types listed above. Assume that the following conditions exist:

1. The notification delay timer is set to three minutes.

2. The root cause is a UCS blade failure, and the final service state for the IaaS service is AT RISK.

If these conditions exist, the following occurs:

1. At zero seconds, event E1 (VMware event for the VM) arrives. The root cause at that time is RC1= VM Failure.

2. A service impact event is sent northbound indicating that the IaaS state = AT RISK, RC=VM.

3. At one minute, event E2 (VMware event for the host) arrives. The root cause at that time is RC2= Host Failure. Since the notification delay timer is set to three minutes, there are no events sent northbound due to the change of root-cause events. Only one minute has passed since the service state change time.

4. At four minutes, event E3 (UCSM event for blade) arrives. The root cause at that time is RC3=UCS blade failure. A service impact event is sent northbound indicating that the IaaS state = AT RISK, RC= Blade.

Figure 3-10 shows the existing capability that Zenoss has to delay notifications and also to send the clearing events. The same capabilities would be extended to the service impact events.

Figure 3-10 Edit Notification Timer

Notification Content

The following fields should be included in the northbound service impact notification:

Folder Name (one up only). The customer name would typically be placed here, but keeping the field generic allows flexibility to use folders in any way desired (e.g., to represent shared infrastructure services, reseller, etc.). The operator can include the option to have a full folder path.

Folder Type. The folder type indicates what the folder represent, e.g., for folders representing the customer name, the folder type would have value the value "customer."

Service Instance Name and systemwide unique ID.

Service Type. This field can be used to filter notifications by type of service that the northbound consumer is interested in, even though each instance of the service may be in different folders which are representing different customers.

Service State. The service state is UP, DOWN, AT RISK, or DEGRADED.

URLs to Service Impact EventDetail. This page acknowledges and closes events and device events.

Timestamp.

Event clearing ID. The ID of the event that is being cleared by this event.

Probable root-cause event name and systemwide unique ID (event with highest confidence level).

Probable root-cause confidence level.

Probable root-cause device, component, and severity.

Impact chain and ID to events in impact chain. The ID can be used to retrieve the impact chain via REST API upon receipt of the notification.

URLs to probable root-cause EventDetail. This page acknowledges and closes events and device events.


Note In CLSA VMDC 3.0, the following fields are not supported: Folder Type, Service Type, and URLs to probable root-cause event detail. In addition, the Event Clearing ID is implemented slightly differently than proposed above. The Service Instance Name & system wide unique ID is implemented in a field called zenImpactUUID. The initial and clearing events have the same zenImpactUUID, however they have states new and cleared.


Root-cause Event Notification

In addition to sending probable root-cause events as part of service impact notification, there is also a need to be able to send only probable root-cause events. For example, in cases of more catastrophic failures where a single root-cause event impacts a larger number of services, northbound systems that are not service focused may prefer to receive only one notification representing the root-cause event and not receive multiple service impacting notifications.

Even in this case, it is desirable to provide the relationship between the root-cause event and the services it impacted. This can be done by including a list of services impacted by the same root-cause event in the root-cause event notification URL or ID.

Root-cause notification is not a separate notification in CLSA VMDC 3.0; instead, the root-cause event is communicated as a field via the service impact notification.

WS or ReST API

The JSON API can be used to obtain the following:

Device model and attributes data

Performance data

Event data

Service data

Most of the information visible via the GUI can also be obtained via the JSON API.

In addition to retrieving data, the JSON API can also be used for the following:

Managing events (acknowledge, clear, close)

Adding devices to be monitored

Setting production state

Initiating discovery and modeling of devices

Managing thresholds

Managing reports

Other configurations

More information on the JSON API can be found at the following URL:

http://community.zenoss.org/community/documentation/official_documentation/api

Northbound Integration Use Case Examples

This section includes typical use cases that illustrate the rich filtering capabilities of the NBI.

Sections

Abstraction via Single Interface

Integration With Multiple Northbound Systems

Abstraction Through Service Overlays

Abstraction via Single Interface

One of the key functions of the SAM layer as defined in CLSA VMDC architecture is the capability to provide a single, normalized NBI that is consistent regardless of the formats of data used by the underlying VMDC components. This allows simplified integration and ongoing interface maintenance with providers existing OSS systems as:

There is only one integration point as opposed to the number of integration points being proportional to the number of VMDC devices and domain managers.

Changes in any of the underlying interfaces on managed devices are absorbed by the SAM as opposed to the provider having to update OSS systems every time there is a change in one of the managed components.

Figure 3-11 illustrates how the VMDC system is abstracted via a single interface to the provider's existing OSS system. The purple areas represent the enhancements for CLSA VMDC 3.0.

Figure 3-11 Single Normalized and Service Abstraction NBI

Integration With Multiple Northbound Systems

This use case example illustrates the need for different types of notifications and northbound filtering capabilities.

Figure 3-12 VMDC CSA Integration in Northbound OSS Systems

In this use case example (Figure 3-12), there is an IT department with two operations teams: one for network operations and one for server/compute operations. In addition, each one of the teams has a ticketing system and a MoM capable of further event processing and RCA. Assume also that the server/compute operations team has an SLA management system used to measure and manage SLA compliance.

Using the extensive filtering capabilities of the northbound notifications, the needs of both of these operations teams and their various northbound systems can be satisfied with a single instance of the service assurance system. In this example, five northbound notification destinations are configured, each with a different filter (also known as a notification trigger) as follows:

All root cause events originated by vCenter or UCSM are sent to the Compute Operations ticketing system.

All service-impact events originated by vCenter or UCSM are sent to the Compute Operations SLA management system.

All other compute events that may require additional analysis are sent to the Compute Operations MoM.

All root cause events originated by network devices are sent to the Network Operations ticketing system.

All other compute events that may require additional analysis are sent to the Network Operations MoM.

Abstraction Through Service Overlays

This use case illustrates the need for service impact notifications from CLSA VMDC. This use case is a prerequisite for integrating CLSA VMDC into CLSA-HCS. To deliver HCS services (voice, voicemail, etc.) to the end customer/tenant, multiple services need to be provided to the customer, which are referred to as service overlays. In a scenario for top-level service such as HCS, there are a number of benefits to only processing abstracted events related to a few underlying services:

Complexity of its fault management system can be reduced significantly if it is only receiving events related to few underlying services (IaaS, MPLS VPN WAN service, etc.) rather than having to deal with device level events from tens of underlying components.

More flexibility to support various business and operational deployment models that vary in which domains and services are owned and operated by the provider offering top-level (e.g., HCS) services.

Figure 3-13 and Figure 3-14 illustrate the service overlay approach for application-based services such as HCS, and the need for service level abstraction from the underlying infrastructure system.

Figure 3-13 HCS Services and Operational Domains

Figure 3-14 Multi-tiered SIA

Performance Management

The following performance management capabilities are provided out-of-the-box in CLSA VMDC:

KPI statistics resource monitoring and trending:

Resource monitoring is partially validated as part of CLSA VMDC.

Performance service impact models for compute and storage:

TCAs utilized as part of SIA

Validated as part of CLSA VMDC

Application response time measurements:

Not validated as part of CLSA VMDC

For details, refer to product documentation on www.zenoss.com.

Performance reporting:

Not validated as part of CLSA VMDC

For details, refer to product documentation on www.zenoss.com.

Dashboards

CLSA VMDC features aggregated SP dashboards, as well as both device level and service level dashboards that operators can use to obtain more details. The following are the key dashboard categories for Zenoss CSA:

Aggregated systemwide resources status dashboards

Service inventory and status dashboards

Infrastructure/resource monitoring dashboards

Event dashboards

Aggregated Systemwide Resources Status Dashboards

These dashboards list all devices with events systemwide, sorted by number of highest priority events, as shown in Figure 3-15.

Figure 3-15 Aggregated Systemwide Resources Status Dashboard

Service Inventory and Status Dashboards

These dashboards show the availability and performance state of all services in the system, as shown in Figure 3-16.

Figure 3-16 Service Inventory and Status Dashboard

Figure 3-17 shows a per-service detailed dashboard, which lists service impact events and related probable root cause events, as well as a visualization of the service model tree.

Figure 3-17 Per-service Detailed Dashboard

Infrastructure/Resource Monitoring Dashboards

These dashboards list the inventory of all devices and their status, as shown in Figure 3-18.

Figure 3-18 Infrastructure Dashboard

Figure 3-19 and Figure 3-20 show a detailed component dashboard and graphical view (example UCS server blade).

Figure 3-19 Detailed Component Dashboard

Figure 3-20 UCS Server Blade Graphical View

Event Dashboards

These dashboards show all events in the console (similar consoles exist per component as well), as shown in Figure 3-21.

Figure 3-21 Event Dashboard

Refer to Dashboards for a use case example of dashboard monitoring.

Reporting

CLSA VMDC provides a range of defined and custom report options, including the following:

Device Reports

Event Reports

Performance Reports

Graph Reports

Multi-Graph Reports

Custom Device Reports

Reports can be exported to external files and systems or can be viewed locally. Reports can also be generated ad hoc or scheduled. Refer to the Zenoss Cloud Service Assurance Installation and Administration Guide for more information.

The following is a list of reports supported out-of-the-box for CLSA VMDC:

Device Reports (9)

All Devices

All Monitored Components

Device Changes

MAC Addresses

Model Collection Age

New Devices

Ping Status Issues

SNMP Status Issues

Software Inventory

Custom Device Reports

Graph Reports

Multi-graph Reports

Event Reports (3)

All EventClasses

All EventMappings

All Heartbeats

Performance Reports (7)

Aggregate Reports

Availability Report

CPU Utilization

Filesystem Util Report

Interface Utilization

Memory Utilization

Threshold Summary

Storage (3)

Clients

Licenses

Disk Firmware

Enterprise Reports (17)

Organizer Graphs

95th Percentile

Defined Thresholds

Interface Volume

Network Topology

Customized Performance Templates

User Event Activity

Notifications and Triggers by Recipient

Datapoints by Collector

Organizer Availability

Maintenance Windows

Interface Utilization

Event Time to Resolution

Data Sources in Use

Users Group Membership

Cisco Inventory

Guest to Datapools

MSExchange (1)

MSExchangeAvailability

VMware (5)

ESXs

VMware Utilization

VMs

Datastores

Clusters

Cisco UCS Reports (2)

Hardware Inventory

Free Slots

Multiservices

This section discusses the CLSA VMDC approach to multitenancy. VMDC architecture supports multitenant delivery, and CLSA VMDC must therefore support an assurance window into these tenant services to equip cloud providers with the ability to assure logically distinct customer services. A related topic, Role-Based Access Control (RBAC), is also presented in this section.

CLSA VMDC Multitenancy

VMDC provides a multitenancy cloud infrastructure by logically separating tenant services that are implemented on a shared physical infrastructure. Tenants consume a portion of network, storage, and compute resources that have been allocated from the larger pool represented by the cloud. CLSA VMDC delivers cloud provider assurance of shared infrastructure devices and their sub-components. In addition, CLSA VMDC supports the multitenancy aspect of the VMDC architecture through the use of defined Tenant Services.

Zenoss CSA enables an administrator to stitch together service element nodes which taken as a whole comprise a specific tenant service. A CLSA VMDC tenant service begins with the creation of the topmost element node named for the tenant. To this tenant node, underlying VMware vSphere and UCS shared infrastructure elements can be discovered and attached. Refer to VMDC Assurance Service Models for more details regarding tenant services.

Using the tenant service modeling feature of CLSA VMDC, cloud customers' services can be assured independently. Elements of the tenant service that are unique to that tenant customer such as specific VM's are visible only to the cloud provider or the service owner. Elements of the service that belong to shared infrastructure, such as a UCS chassis or a storage device are visible across multiple tenant services, as would be expected. In fact, if a shared device experiences a fault condition, all services associated with that device should be impacted, however, any fault condition associated with unique elements of a tenant service are not visible to other tenants.


Note This phase of CLSA VMDC only supports cloud provider visibility into dashboards and service impact trees. Tenant customer visibility into service impact trees via customer portals will be supported in future releases.


RBAC Implementation

As a cloud providers' infrastructure increases in scale, it becomes important to provide a segmentation of operations capability to implement a division of responsibility. CLSA VMDC fulfills this need with its RBAC implementation. Beyond this division of responsibility capability, RBAC can also be used to support groups of users with limited visibility into specific tenant services. Table 3-5 lists the out-of-the-box roles that may be used to segment cloud provider operations responsibilities and access.

Table 3-5 Global User Role Definitions 

Role
Definition

ZenUser

Provides global read-only access to system objects.

ZenManager

Provides global read-write access to system objects.

Manager

Provides global read-write access to system objects and read-write access to the Zope object database (which includes all devices, users, and event mappings)

ZenOperator

Provides users the ability to manage events, i.e. acknowledge, move to archive, etc.


These predefined roles are global in scope, such that an operator may access all cloud objects, but only be allowed certain operations. In addition to these global user roles, users may be defined that are more limited in scope. These user roles may be assigned to organizational groups to manage a subset of the entire infrastructure or even specific tenant services.

Organizational groups are used to collect subsets of infrastructure and/or services into logical categories for segmented operations. Table 3-6 lists the broad group categories and suggested uses for each group type. Groups can be devised for each tenant such that a customer's tenant services can be assigned to a tenant group.

Table 3-6 Device and Service Group Categories

Group Categories
Group Purpose

Group

Can be used for collecting similar devices, e.g. all switches group, all compute devices group

Systems

Can be used to collect all equipment with a specific data center, e.g. data center A, data center B

Locations

Can be used to collect devices by geographic boundaries, e.g. city, state, or even specific device rack


Figure 3-22 illustrates a list of users with either global or customized group role assignments. Users with the global user roles would belong to the cloud provider.

Figure 3-22 Custom User Groups

Users with non-global roles can belong to either the cloud provider or even a cloud customer. While RBAC is only supported for the cloud provider in this phase, it could be the mechanism to deliver a cloud customer, or multiservices, assurance portal in future phases.