sysbench-developers team mailing list archive

Thread
Date

Re: [Question #254277]: difficulty interpreting results of disk i/o test

To: sysbench-developers@xxxxxxxxxxxxxxxxxxx
From: Alexey Kopytov <question254277@xxxxxxxxxxxxxxxxxxxxx>
Date: Wed, 10 Sep 2014 06:23:19 -0000
Reply-to: question254277@xxxxxxxxxxxxxxxxxxxxx
Sender: bounces@xxxxxxxxxxxxx

Question #254277 on sysbench changed:
https://answers.launchpad.net/sysbench/+question/254277

    Status: Open => Answered

Alexey Kopytov proposed the following answer:
I guess the difference between prepare and run is that prepare creates a
new file, while 'run' with --file-test-mode=rndrw generates a mixed
read/write workload, which also does cached reads by default. So I/O
performance between the systems is amortized by the cache.

Things to try:

--file-test-mode=rndwr (i.e. a write-only workload)
or
--file-test-mode=rndrw --file-extra-flags=direct (i.e. a mixed read-write workload, but with OS caching disabled via O_DIRECT).

-- 
You received this question notification because you are a member of
sysbench-developers, which is an answer contact for sysbench.