Paisit Blog

SWAT - Sun StorageTek Workload Analysis Tool, is a powerful data analysis package, designed to facilitate the analysis and subsequent resolution of storage hardware performance problems. SWAT is a host-based (support Windows and Solaris), storage-centric performance tool. It analysis I/O from host perspective (that's where application running and SLA is measured from). SWAT is written by Sun engineer - Henk Vandenbergh.

SWAT have superior user interface (GUI) that allow user can easily understand what's going on in their I/O in system in various aspects. For example, response time, IOPS, MB/s, read/write ratio, queue depth, I/O distribution, block size and son. These information can help user to determine what's the bottleneck and potentially how to solve it by leveraging these information properly. For example, if read/write ratio shown that majority is write and user is suffering on response time, potentially change RAID configuration from RAID5 to RAID1 would help. Another example is that, if workload distribution is not balance to volume/LUNs and those volume/LUNs are not sitting on the same layout (disk group), potentially re-layout (by make one big disk group and put these LUNs on top) can leverage resource better and lead to have better response time (Share nothing v.s. Share everything concept).

Although SWAT is now only available on Solaris and Windows platform (run as agent). The latest version - 3.0 allow us to import OS I/O performance data from other platforms (i.e.,AIX, Linux) to do high-level analysis (no agent running).

SWAT uses two methods to capture data regarding storage performance: 

• The Swat Performance Monitor (SPM) generates a long term, high-level view of performance in both the Solaris and Microsoft Windows operating environments. 

• The Swat Trace Facility (STF) generates a short term, detailed view of storage activity in both the Solaris and Microsoft Windows operating environments. Swat is a storage performance-monitoring tool; thus it does not collect information about server processes, context switching, file systems, or network connections. Nor does Swat profile an application, but it does capture the impact of an application’s demand for I/O on a storage system.

Which method above the user can use?. It's really depend on how much information they want to know. STF is low level analysis but a lot of raw data generated. While SPM is high level analysis which can not give user in depth data but much lesser raw data. Usually, for user who don't know what's the right period to do data capturing, it's recommended to run SWAT in SPM mode to see what's the period within a day they are interested and run again with STF mode to zoom down the detail (1 hour period).

SWAT is not only doing I/O analysis but also can work & integrate with other tools very well such as  VDBENCH tool. VDBENCH is the I/O workload simulation tool. SWAT can generate input (as a replay file) to VDBENCH tool for real workload simulation purpose. By running SWAT in STF mode (or unix script), it can re-generate the replay file that represent exact I/O activity in production system. This way will give user the great flexibility to test the storage configuration based on the real I/O behavior data without having real application running.