----- Message d'origine -----
De: "redhat@mckerrs.net" <redhat@mckerrs.net>
Date: Lundi, Novembre 12, 2007 10:28 pm
Objet: Re: Re : Re: [CentOS] Need advice on storage
À: CentOS mailing list <centos@centos.org>
>
>
> Guy,
>
> how many drives can the server take ? and how many are there in
> it just now ?
>
> I'd even go as far as to say install the OS and the data on the
> 3 x 320gb sata drives and ditch the 80gb IDE. That is what I do
> and it is one of the strengths of software raid, the fact that
> you can mix and match raid types on the same physical disks. I'd
> go for a triple mirror for /boot and install the rest of the OS
> on the RAID-5 set or along with the data (grub can't boot of
> raid-5 as far as I'm aware).
>
>
> Alternatively if re-installing or copying your current OS onto
> new drives is not practical at this point I'd get 2 x 320g
> drives mirror them and also get that extra 1gb of RAM. The extra
> RAM could make more of a difference than 10k rpm drives. I dont
> think that you'll get anywhere near the performance you'd expect
> from a stripe compared to a mirror. As mentioned earlier,
> Raptors, IMO, are overrated and they are way hotter and noisier
> due to their higher spindle speed.
>
> How is the box performing ? what are you noticing when it is
> slow ? Wait IO ? Can you determine if it is IO bound or CPU
> bound ?
>
>
> Cheers,
>
> Brian.
>
The server can physically take 5 drives.
As for RAM, suggestion taken. I'll go buy 1 more Gig. I know that Linux is aggressive when using RAM / Cache so i can just help.
When all clients are connected and when they seem to refresh their mail cache and office activity is high, %iowait goes up to 45-55. I have a script that check and log lsof used by Communigate (every 5 minutes). Sometimes, i see lsof as high as 130 000. I had to raise a couple of parameter (in sysctl.conf and limits.conf)because we had Communigate disconnecting clients and i saw messages like:
IMAP the 'accept' call failed. Error Code=too many files open in this process
I'll probably start another thread for this problem. The default settings of CentOS for open files were very low. I had to raise the values and now i'm at 300 000 and i Communigate still display errors from time to time even if my not so precise lsof logging script never shows more than 130 000. I had even "too many files" errors with lsof log showing as low as 65 000. I'm a little confused. Maybe i'll need more elaborate debugging tools.
As you said, i'm in a case where migrating system drive to a RAID 5 array wouldn't be easy.
But i have a new possible solution that arosed when i went to see the physical system tonight: There is a Promise VTRAK 15100 SCSI enclosure in the rack with a free SCSI bus and space to put new SATA drives inside. I think we have a spare Adaptec 39160 so tomorrow, i'll check that (it's a PCI-X card but i think it should work PCI too). If i have the Adaptec, then i'll probably buy 4 x Western-Digital 250 Gigs "YS" (RAID serie) @ about 75$ each, put them inside the VTRAK and configure them in RAID 10 for a total of 500 Gigs. I briefly checked the VTRAK documentation and they were talking about "up to 200 MBps" throughput and support for NCQ. It remains to be seems how it could perform with 4 disks in RAID 10. But it should easily beat a single 200 Gigs IDE drive!
For now, i did hdparm -t /dev/hdc (stoopid IDE mail drive!) and it gave me 38.21 MB/sec.
Thanks!
Guy Boisvert