Stress testing your machine -- what program?

[William Park]

On Tue, Apr 19, 2011 at 10:56:09AM -0400, Lennart Sorensen wrote:
> On Tue, Apr 19, 2011 at 06:25:27AM -0400, Anthony de Boer wrote:
> > Just a thing to check, but make sure that RAID array isn't running in
> > degraded mode.  If one of the disks failed, or it thinks the array is
> > out of sync with itself, then everything you would have read from the
> > failed disk has to be reconstructed from the other three, and
> > performance goes to lunch.  It's been awhile since I used software RAID
> > on Linux, but /proc/mdstat should be the file that shows you the status.

It did happened to me once, and disks were thrashing a lot.  In fact,
that was what caused me to look at /proc/mdstat, and realized that it
was in the middle of re-sync.

> 
> And make sure you told the filesystem block allocator about the raid
> layout so that it doesn't cause unnecesary read/modify/write cycles on
> the raid5.

Well, I used 64kB chunk size, as in the following:

    # chunksize = 64kB
    # blocksize = 4kB
    # stride = chunksize / blocksize = 64 / 4 = 16
    # (Raid5) stripe-width = stride * (N - 1) = 16 * (4 - 1) = 48
    # (Raid0) stripe-width = stride * (N) = 16 * (4) = 64
    #
    mkfs.ext4 -b 4096 -m 0.1 -E stride=16,stripe-width=48 /dev/md0

Are you saying I should use different chunk size?
-- 
William
--
The Toronto Linux Users Group.      Meetings: http://gtalug.org/
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://gtalug.org/wiki/Mailing_lists