Storage Test FAQs
How do you know these problems are associated with disk issues?
After investigation, the root cause of this set of problem reports was found to be storage devices with latency issues (long IOs). Changing the device or updating the device’s firmware led to a significantly improved user experience.
By playing back user traces associated with this category of problem report across a range of drives, IO patterns associated with the latency issue were identified. The WinSAT write flush policy tests were then built as a small set of mixed read/write tests which could run quickly and identify devices with those latency issues.
What do the write flush policy tests do?
A combination of sequential and random IOs (reads and writes) that can trigger the long-running IOs, depending on the disk’s caching policies (specifically the write flush policies).
How do you know the write flush policy tests are meaningful?
Microsoft ran many tests across a variety of workloads, analyzed the results, compared the results to those seen using original problem traces, and kept the sub-set of tests which most consistently identify the same disks flagged by the original problem traces.
How do the write flush policy tests impact WinEI scores?
1. If a pattern of long IOs are detected, the WinEI Disk score is capped. In Beta, there are two levels of capping, at 1.9 and 2.9.
- A cap of 1.9 corresponds to having either
- IOs greater than 600 milliseconds, or
- mean IO >= 22milliseconds and 95th percentile IO >= 40 milliseconds
- A cap of 2.9 corresponds to either
- IOs greater than 520 millseconds
- Mean IO >= 11 milliseconds and 95th percentile IO >= 33 milliseconds
2. Since the overall Windows Experience Index is based on the lowest of the 5 sub-scores (one each for Disk, CPU, Memory, Desktop Graphics and Gaming Graphics), if the Disk score is low it can cause the overall WinEI score for that system to be low.
For devices without latency issues, how are the higher scores defined?
The overall WinEI scale is 1.0 to 7.9. In the Beta release we are using the following scoring :
-
5.0 to 5.9 represents disks with a good sequential read performance and a good write flush policy. Most mechanical drives are expected to show performance that is rated under 5.9.
-
6.0 to 6.9 represents the expected performance for storage devices that have very good random I/O performance.
- Note that most currently shipping devices in this class score in the 6.0 to 6.5 range. As faster devices reach the market, they can achieve 6.6 and higher. -
7.0 to 7.3 maps to the best performing, currently shipping devices in terms of sequential and random I/O performance.
- We expect that over the 2010 and 2011 time frame, drives will dramatically increase in performance, so have left plenty of room in the 7.0 range for these devices.
- Devices in the 7.0 range can sustain high rates of both random and sequential I/O.
Support Tech ARP!
If you like our work, you can help support out work by visiting our sponsors, participate in the Tech ARP Forums, or even donate to our fund. Any help you can render is greatly appreciated!
<<< Write Flush Policy Tests, Assessing Enhanced Storage Systems : Previous Page | Next Page : Storage Test FAQs (Continued) >>>