Comments on: On Benchmarks on SSD http://www.mysqlperformanceblog.com/2010/07/14/on-benchmarks-on-ssd/ Percona's MySQL & InnoDB performance and scalability blog Sat, 11 Feb 2012 16:45:54 +0000 hourly 1 http://wordpress.org/?v=3.2.1 By: Shirish Jamthe http://www.mysqlperformanceblog.com/2010/07/14/on-benchmarks-on-ssd/comment-page-1/#comment-770788 Shirish Jamthe Fri, 06 Aug 2010 02:15:19 +0000 http://www.mysqlperformanceblog.com/?p=3353#comment-770788 Patrick, Be careful of the loopback device. In our testing it shows that it lies about fsync, so the process that calls fsync on loopback device file may think data is persistent but that is not usually the case. Patrick,

Be careful of the loopback device. In our testing it shows that it lies about fsync, so the process that calls fsync on loopback device file may think data is persistent but that is not usually the case.

]]> By: peter http://www.mysqlperformanceblog.com/2010/07/14/on-benchmarks-on-ssd/comment-page-1/#comment-769460 peter Fri, 16 Jul 2010 16:40:24 +0000 http://www.mysqlperformanceblog.com/?p=3353#comment-769460 Patrick, I would be careful with approaches like this. You do not really know what other support from OS and filesystem for SSD devices may be disabled... plus With reads/writes sometimes at 1GB/sec rate the overhead on CPU/Memory bus from extra copying may not be trivial. I think it is practical to run benchmarks as close to production setups as possible for them to have highest relevance Patrick,

I would be careful with approaches like this. You do not really know what other support from OS and filesystem for SSD devices may be disabled… plus With reads/writes sometimes at 1GB/sec rate the overhead on CPU/Memory bus from extra copying may not be trivial.

I think it is practical to run benchmarks as close to production setups as possible for them to have highest relevance

]]> By: Patrick Domack http://www.mysqlperformanceblog.com/2010/07/14/on-benchmarks-on-ssd/comment-page-1/#comment-769386 Patrick Domack Thu, 15 Jul 2010 13:20:10 +0000 http://www.mysqlperformanceblog.com/?p=3353#comment-769386 Wouldn't mounting the device via the loopback device hide the fact it's an ssd from the os, making the trim command not work. Then your one step closer to better results, at the expense of some more cpu overhead. Wouldn’t mounting the device via the loopback device hide the fact it’s an ssd from the os, making the trim command not work. Then your one step closer to better results, at the expense of some more cpu overhead.

]]>