IN THIS EDITION
Introduction The STB Suites Disk Manufacturing Module (DMM) has the ability to drive I/O to and from disk drives at the disks full or fastest I/O rates. In addition it collects highly detailed performance metric information in an individual log file for each drive under test. These features allow detailed comparison and analysis of disk drive performance. Seeing the performance of a single drive is simple and as we’ll show in this article comparing the performance of multiple drives is just a matter of using a few common tools. Log file text metrics Here is an example of two log files created during one test session: Graphing the log file metrics Graphing the pertinent performance information proves that old adage to be true, that a picture is worth a thousand words. Here we see four drives comparing the Write, Read, and Verify transfer rates: Graphing data from DMM log files
The example graphs above were generated using the Microsoft Excel spreadsheet. The data used to generate both of these graphs is simply this:
Q.“Can you explain what all the IO profiles in SCSI Toolbox accomplish, what exactly do they do?” A. Streaming Test: Access is sequential, transfers are from 64-to-128 blocks, 20% of the I/O are writes, 80% of the I/O are reads OLTP Test: Access is random, transfers are from 4-to-32 blocks, 20% of the I/O are writes, 80% of the I/O are reads FileServer Test: Access is random, transfers are from 8-to-128 blocks, 20% of the I/O are writes, 80% of the I/O are reads WebServer Test: Access is random, transfers are from 1-to-1024 blocks, 100% of the I/O are reads Workstation Test: Access is a combination of sequential & random with 80% of the accesses are random and 20% are sequential, transfers are from 1-to-32 blocks, 20% of the I/O are writes, 80% of the I/O are reads The Workstation Test is quite complex in terms of how it accesses a drive. 80% of the time when we generate the next LBA for an I/O that LBA is picked randomly. BUT 20% of the time the next LBA is sequential, which means if the previous I/O hit LBA X, then the next LBA is X + ”trans len” (so that the 2 I/Os are “next to each other” just as sequential does). For concreteness sake, let's say each I/O did a transfer of 10 blocks. Then the LBAs generated would be something like 781296, 1356, 1366, 1000225, 82, 987123, 101023, 101033, 15, 23567891 The above is 10 I/Os with the start LBA listed; note that exactly 2 I/Os have start LBA exactly 10 more than the previous (i.e. 1356 & 1366, and 101023 & 101033, all the other 8 I/Os are randomly generated)
You may have read from our July newsletter that STB is offering some really neat new server testing functionality. (See full release notes) We're excited to roll out our 9.0 STB Suite release. However, we have delayed the 9.0 release to allow for more extensive testing. If you're interested in becoming a beta tester for the 9.0 STB Suite release, please make sure you meet the following requirements then contact STB.
If you're interested in the beta, please use the contact form or call sales at 720.249.2641.
Here is a list of some recent customer training sessions that STB has conducted - live, interactive web sessions presented by STB programmers:
Contact Jeremy Wolfe at (720) 249-2641 today to schedule your own custom training session!
|
|