In today's electronic economy, enterprises are facing an expanding set of data-protection challenges and regulatory data-retention requirements. The IBM Tivoli Storage Manager (TSM) family of products is designed to provide centralized, automated data protection. IBM Tivoli Storage Manager enables you to protect your organization's data from failures and other errors by storing backup, archival, and compliance and disaster-recovery data in a hierarchy of offline storage.
Limited bandwidth and latency are challenges created by the WAN in a centralized deployment of any application, including IBM TSM. The Cisco® Wide Area Application Services (WAAS) solution uses industry-leading data redundancy elimination (DRE), data compression, and TCP Flow Optimization (TFO) to reduce the communications wait time and overall time needed to back up remote file servers, optimizing the performance and end-user experience of Tivoli Storage Manager over the WAN. This document reviews demonstrated performance improvement metrics of Tivoli Storage Manager using Cisco WAAS over the WAN.
Cisco WAAS Overview
Cisco WAAS is a multilayer application acceleration and WAN optimization solution that improves application performance over the WAN, thus enabling application and server centralization into fewer data centers. Acceleration for backup applications and the associated protocols is achieved through the following Cisco WAAS optimization features:
• Transport Flow Optimizations: TFO provides standards-based, field-proven throughput improvements for TCP-based applications while maintaining packet-network friendliness and safe coexistence with other network nodes communicating using standard TCP implementations. TFO terminates TCP sessions locally and transparently optimizes flows that traverse the WAN, thereby shielding communicating nodes from WAN conditions.
• Data Redundancy Elimination (DRE): DRE is an advanced form of network compression that allows Cisco WAAS to maintain a database of data that has been seen previously traversing the network. This information is used to remove redundant transmission patterns so that they do not have to use the network. For repeated patterns, only instructions need to be sent, and the original message is rebuilt in its entirety by the distant appliance. This strategy enables significant levels of compression and helps ensure message and application coherency in that the original message is always rebuilt and verified by the distant Cisco Wide Area Application Engine (WAE). Because DRE is application agnostic and bidirectional, it is effective regardless of the direction of traffic flow. Thus, data patterns that have been identified for one application protocol can be used by other applications, and patterns that have been identified for one direction of traffic flow can be used to remove redundancy for traffic flowing in a different direction.
• Persistent Lempel-Ziv (LZ) compression: Persistent LZ compression is a standards-based compression with a long-lived connection-oriented compression history that can be employed to further minimize the amount of bandwidth consumed by a TCP flow. LZ compression can be used in conjunction with DRE or independently. LZ compression can provide from 2:1 to 5:1 compression depending on the application being used and data being transmitted. It is especially helpful for data that has not been previously seen and suppressed by DRE.
• Web Caching Communications Protocol (WCCP): Using built-in WCCP in routers and switches based on the Cisco IOS® Software, client traffic is transparently redirected to a Cisco WAE. This redirection enables the Cisco WAE to transparently apply optimizations to application traffic, such as file requests. For protocols such as Common Internet File System (CIFS) where traffic localization can be applied, Cisco WAAS can dramatically reduce transmission costs and download times.
WCCP enables clustering of a series of Cisco WAE appliances for high availability, load-sharing, and scalability, called a Cisco WAE service group, to interact with a single or multiple routers. With these clustering capabilities, network administrators can easily scale their Cisco WAE deployment to handle heavy traffic loads, high availability, and failover scenarios. Cisco clustering technology enables each Cisco WAE member to work in parallel, resulting in linear scalability. Up to 32 Cisco WAEs can be clustered in one service group to scale to the desired capacity.
Although WCCP offers the most available and scalable solution for WAN optimization, deployments of Tivoli Storage Manager can also be deployed inline, with all traffic sent through the Cisco WAE with TCP traffic optimized and other traffic passed through. This mechanism provides a simpler deployment option for smaller deployments.
Test Methodology and Topology
Figure 1 shows a centralized Tivoli Storage Manager deployment with Cisco WAAS. Each edge site was configured with a Cisco WAE-512 appliance and a Cisco 2811 Integrated Services Router with 384,000 Frame Relay connections. The core was configured with a Cisco WAE-512 appliance and a Cisco Catalyst® 6509 Switch with a Supervisor 720-3A connected to the WAN with a T1 Frame Relay circuit.
Testing was done in a customer's production environment. WCCP redirection was used at all sites to gain maximum availability and scalability. WCCP also allowed transparent installation of the solution with no network downtime. Each remote site was running Windows 2000 Servers with IBM TSM Storage Manager Version 5.3.1.6 backup archive client. Latency measurements were gathered by ping results that were run during nonproduction hours. The IBM server graph was used to measure session and communication wait time for the IBM TSM application.
Figure 1. WAAS Topology
Figure 2 shows the reduced time to complete an IBM Tivoli Storage Manager backup using Cisco WAAS. Site 1 experienced 29 milliseconds (ms) of latency, and Site 2 experienced only 14 ms. As the latency increases, Cisco WAAS uses TFO and DRE to aid in throughput performance, reducing the amount of time needed to back up file servers.
With Cisco WAAS enabled, the time needed to back up Site 1 was drastically decreased, from 5 hours and 22 minutes to 1 hour and 17 minutes. The other sites tested followed this trend and also had significantly decreased backup times.
Figure 2. Session times
Figure 3 shows the improved data rate performance of IBM Tivoli Storage Manager using Cisco WAAS. Site 1 had 29 ms of latency, and Site 2 had only 14 ms. IBM TSM is very latency sensitive because of the time it uses for acknowledgments, which is illustrated in Figure 2 by the communication wait time.
With Cisco WAAS enabled, Site 1 had a 5X data rate increase, from 30.8 Kbps to 151.3 Kbps. A 2X increase was achieved at Site 2.
Figure 3. Session Data Rates
Figure 4 shows a bandwidth graph for Site 1 over a 1-week period. The green spikes show the bandwidth that the backup uses each night. A reduction in bandwidth was seen after installing the Cisco WAE, as shown by the vertical red line on Monday night.
Figure 4. Site 1 Bandwidth Utilization over Time with and without WAAS
Figure 5 show a bandwidth graph for Site 2. The taller green spikes at night show the backup traffic. After Cisco WAAS was installed, shown by the vertical red line, a significant amount of traffic was reduced on this link.
Figure 5. Site 2 Bandwidth Utilization over Time with and without WAAS
Conclusion
Organizations are faced with the challenge of maintaining data and tape management in remote locations. Centralized backup over the WAN is an approach to reduce maintenance and hardware cost by centrally locating tape and disk resources. Latency in the WAN has adverse affects on backup applications, which Cisco WAAS can help overcome.
Cisco WAAS has demonstrated performance improvements of up to 5X for IBM Tivoli Storage Manager. With higher WAN latency, Cisco WAAS shows even greater performance improvement and WAN bandwidth reduction. By using Cisco WAAS in the enterprise data center as part of a central backup strategy, customers can increase backup performance while freeing bandwidth for interactive users.
