[Gluster-devel] Re: Bonnie++ bailing out

[Samuel Douglas]

On Sun, Sep 21, 2008 at 7:52 AM, KwangErn Liew <ke.liew at gmail.com> wrote:
> On Sat, Sep 20, 2008 at 9:19 PM, Krishna Srinivas <krishna at zresearch.com>wrote:
>
>> KwangErn,
>> Did increasing transport-timeout help?
>
>
> Am still waiting for it to complete. Been running more than one day now. So
> far so good. :)
>

When testing GlusterFS earlier in the year, I found that the size
recommendations in the Bonnie documentation (at least twice your RAM
size to prevent the kernel's block cache from affecting the benchmark
results) were unnecessary on NFS and FUSE filesystems. We had 4GB of
RAM in our nodes, so this was a good discovery. Changing the benchmark
size down to a couple of hundred megabytes did not seem to affect the
results at all, just the amount of time it took to get them.

The settings for the read-ahead and write-behind translators greatly
affect the times for the random access benchmarks.

~~Sam