Re: Consistent throughput challenge -- fragmentation?

[Brian Cain <brian.cain@xxxxxxxxx>]

On Mon, Feb 25, 2013 at 4:18 PM, Brian Cain <brian.cain@xxxxxxxxx> wrote:

Here's the info from the FAQ, I'll bundle up the benchmarks.

On Mon, Feb 25, 2013 at 4:16 PM, Dave Chinner <david@xxxxxxxxxxxxx> wrote:

Post your benchmark script, along with the results you see, and all

the other information listed here:

http://xfs.org/index.php/XFS_FAQ#Q:_What_information_should_I_include_when_reporting_a_problem.3F

Attached is the script which runs the benchmark.  It refers to a "/usr/bin/write-bench" which I will work on seeing how I can share.  But it does little more than allocate a buffer, read into it from /dev/urandom, then "open(O_DIRECT)/write(16MB)/write(16MB)/.../write(16MB)/close()" and measure the duration of the write() calls using gettimeofday() or clock_gettime(CLOCK_MONOTONIC).

--
-Brian

wb-model
Description: Binary data