Redundant NFS storage setup (part 3) : The disappointing PERC 5/E (solved?)

[Matthias Saou]

Matthias Saou wrote :

[...]
> One thing I've just realized which puzzles me a little is this :
> 
> # hdparm -t /dev/sdb
> /dev/sdb:
>  Timing buffered disk reads:  1876 MB in  3.00 seconds = 624.76 MB/sec
> # hdparm -t /dev/sdb1
> /dev/sdb1:
>  Timing buffered disk reads:  872 MB in  3.00 seconds = 290.66 MB/sec
> 
> The speeds vary slightly from one run to the other, but the difference
> between the entire block device and the partition I've created above is
> a factor 2. With sdb1 starting at 17.4kB, same thing.
> 
> Could this be a result of the unalignment? I can't help but think it
> actually could... I just tested the same against sda and sda1 (RAID1 of
> 15k SAS drives on the internal SAS5i), and both give me the same speed.

I just tested again with the partition starting at the very beginning
(17.4kB) and 128kB, both give 280-300MB/s depending on the run. As a
wild guess, I just tried starting at 64kB and got 550-560MB/s with
hdparm, while I have my RAID5 configured with 128kB stripe size...

With iozone on that partition, I see no significant changes, but my
"real world test" of sequential reads now manages to get 480-550MB/s
sustained for 3 non cached 10GB files, which is the highest I've ever
seen with this setup.

But really, why starting the partition at 64kB gives me the best results
while my RAID5 has a 128kB stripe size is beyond me. I'm glad I tried
it, though ;-)

Matthias

-- 
Clean custom Red Hat Linux rpm packages : http://freshrpms.net/
Fedora release 8 (Werewolf) - Linux kernel 2.6.23.9-85.fc8
Load : 1.71 1.12 0.66