We recently had an opportunity to test IBM Tivoli Storage Manager (TSM) Client to a TSM Server in our HyperIP lab. When doing any kind of application verification or performance testing it is important to first determine the overall limits of the native application with and without WAN acceleration.
Lab testing in an emulated environment is a good way to test applications because you can mimic certain network topologies and characteristics. In our case the HyperIP lab consists of two HyperIP WAN Optimization virtual appliances, two windows servers, and a distance simulator for the WAN. The simulator has the ability to inject packet loss, network latency and other network conditions over various bandwidths that can degrade application performance.
The main objective with any test is to try to validate whether the HyperIP can accelerate the application over various distances with varying latency and packet loss scenarios. Every application has its own performance characteristics and limitations. The same is true for WAN networks. They are about as unique as fingerprints.
Like many backup applications TSM was designed for the data center and performs very well when moving data short distances. Since we are truly becoming a global society is it important to be able to move data over longer distances which is clearly a requirement of cloud storage environments.
With the case of IBM TSM, we started off testing with a simple delay of 10 ms round trip time (RTT). At this relatively short distance TSM slowed by 80% compared to its native performance. This is typical application degradation due primarily to the inefficiencies of the TCP transport and not necessarily the fault of the TSM application. When HyperIP was added to the configuration, the TSM application was able to achieve throughput equivalent to native performance and no delay. In fact HyperIP was able to help TSM achieve near native performance rates at distances represented by 40 ms RTT, 80 ms RTT, 320 ms RTT all the way up to a 1 second RTT. This is a testament to how well TSM and HyperIP interoperate together.
Many applications have internal limitations such as outstanding operations, queue size, or queue depth that artificially restrict the application’s ability to maximize throughput. That was certainly not the case with TSM. TSM can certainly pump data over the network when it is not encumbered with TCP performance issues. When operating TSM with HyperIP, the two combined can sustain the same throughput rates whether running across town, across the ocean, or around the world. That was very impressive. TSM over HyperIP brings LAN-like performance to WAN-based remote backups.