Dell Poweredge 2950 with PERC 5/i RAID

[Claudio Nasuti]

On Tuesday 10 October 2006 11:06, Johan De Meersman wrote:
> Giulio Orsero wrote:
> > On 10/9/06, Claudio Nasuti <claudio.nasuti at acsys.it> wrote:
> >   
> >> The installation went smooth like a breeze but when I made a small IO test I have been shocked
> >> IO seems very poor as everyone can figure by numbers
> >> [root at adm-sdf1 ~]# time dd if=/dev/zero of=./DummyFile bs=65536 count=65536
> >> 65536+0 records in
> >> 65536+0 records out
> >>
> >> real    1m49.236s
> >>     
> >
> > If I'm not mistaken the test shows you get about 37MB/s.
> > If this is very poor, what numbers did you expect?
> >   
> 
> Considering I'd expect somewhere between 50 and 70 MB/s sequential write
> throughput on servers with 72G 15krpm disks in raid 5, I think this is
> reasonable, considering the slower and larger disks.
> 
> Just did the same test on a 4 year old system, and I got about 45MB/s.
> 
> Remember that write throughput depends on rotation speed, number of
> spindles and stripe size - if your stripe size is identical to your
> transaction size, every transaction will be handled by a different
> spindle of the disk array, optimizing parallelism.
> 
> If you're running 2.6 series, you can also have a look at using the
> deadline I/O scheduler instead of the standard (anticipatory) one -
> anticipatory assumes only a single spindle, while deadline can take
> advantage of command tagging and queueing. YMMV.
> 
> 
> 
> 
> -- 
> Think twice before speaking, but don't say "think think click click".
I found the way even if it is blamed as dangerous
I installed the Dell Open  Manage and I configured on storage section
under virtual disk a different policy. 
I changed the policy from "write through" to "write back" setting and my small test on write
shows:[root at adm-sdf2 ~]# time dd if=/dev/zero of=./DummyFile bs=65536 count=65536
65536+0 records in
65536+0 records out

real    0m29.481s
user    0m0.028s
sys     0m7.627s
which means 136.5 MB/s.
I'm aware that I can loose but for my applications performace gain is more relevant

Claudio
 

----------------------------------------------------------------------
ATTENZIONE: le informazioni contenute in questo messaggio sono da considerarsi confidenziali ed il loro utilizzo e' riservato unicamente al destinatario sopra indicato. Chi dovesse ricevere questo messaggio per errore e' tenuto ad informare il mittente ed a rimuoverlo definitivamente da ogni supporto elettronico o cartaceo. 

WARNING: This message contains confidential and/or proprietary information which may be subject to privilege or immunity and which is intended for use of its addressee only. Should you receive this message in error, you are kindly requested to inform the sender and to definitively remove it from any paper or electronic format.
----------------------------------------------------------------------