I just tested the new Intel X25-E Solid State Drive on RHEL-4 64-bit with three file systems (ext2, ext3 and XFS) using the sysbench fileio test.
The test is done to gauge the performance of SSD as a temporary solution to act as tmpdir for MySQL server where result set temporary files are heavily created.
Here is the test results:
Overall, ext2 yields much better results on random IO along with sequential read. All file systems are mounted with noatime and XFS is mounted with noatime + nobarrier
So, even if one decides to use SSD as MySQL data storage; it will be nice to avoid it for any log storage (error log, slow and general query logs, bin-logs and innodb transaction log) due to slower sequential writes. Sequential write on ext2 with O_DIRECT seems to be better than any other file system.
New blog post: http://tinyurl.com/dyrgqv – SSD Peformance With Different File Systems
Interesting, I was using XFS all these days, and never experimented with ext2
Thanks for sharing this Venu. Lots of interest in MySQL with SSD these days but not a lot of benchmark data yet.
SSD 上居然應該用 ext2… # http://tinyurl.com/dyrgqv
說 http://tinyurl.com/dyrgqv (SSD應使用無Journal的Ext2) http://plurk.com/p/jmf87
This result is different when HDD use?
or same tendency?
here is SSD benchmark with using MySQL.
http://d.hatena.ne.jp/sh2/20090325
Hi Venu,
It is not clear from the barchart if the
vertical axis is throughput (higher numbers
are better) or latency (lower numbers are
better). Please advise.
deepblue