I'm going to go out on a limb and presume that you're using IDE. By the
nature of IDE's architecture, it can truly only serve one disk request at a
time, so adding IDE drives won't improve performance that much. By moving
to a SCSI architecture, you could have multiple disks performing
simultaneous read/write requests. If you were to go with a RAID0 SCSI
setup, your disk IO would be tremendous, but you would have no type of
failover; however, in light of the fact that you have none now... this
doesn't look too bad.
RAID5 would be your best bet from a performance/reliability standpoint, but
it will be expensive. To reach the capacity of the 200GB you are currently
using, you would require 3 146GB drives, ending up with approximately 292 GB
of disk space. Your write performance will not be nearly as good as it is
with RAID0, but your read performance will still be quite fast and you will
have an added level of redundancy.
As well, you might check to see if your RAM is capable of CAS 2. If so,
ensure that it is specified as 2 in your BIOS and not 3. In a
quad-processor architecture, this could make a noticable difference. If it
is only capable of CAS3, you might consider upgrading to RAM that is capable
of CAS 2.
These changes will be transparent to your UML host/client configuration(s),
excepting that the performance could be increased drastically, depending
upon what you mean by IO. Have you gathered statistics on the server over a
given time period, (day, week, month), so that you know how and when the
load is shifting? I'd recommend monitoring your CPU usage, memory usage,
disk IO (reads/writes per second) and network IO just for kicks. A simple
MRTG setup should work well.
Feel free to eMail me if you would like leads on hardware or more hardware
advice. I'm no UML expert, but I understand hardware performance. :)
----- Original Message -----
From: "Shawn J. Wallace" <shawn@...>
Sent: Friday, March 12, 2004 12:36 PM
Subject: [uml-user] Improving I/O performance
> I am currently running a redhat machine with around 12 UML's on it, and I
> notice that at times the I/O really suffers.
> The hardware is a quad-xeon with 2GB RAM and a 200 GB hard drive, using a
> custom kernel with the SKAS patch. Currently each UML instance is using
> the virtual storage (file-based, COW, etc).
> I'm looking to improve the performance if I can, and I was thinking that
> I add another drive to it I could partition it up to have a partition for
> each UML and access it directly. Would this improve my I/O
> performance? If so, any ideas on what kind of gains I might have?
> Any other tips for improving performance?
> Shawn J. Wallace <shawn@...>
> Justweb Inc. -- http://www.justweb.com
> Tel: (519)652-6599 Fax: (519)488-1045
> This SF.Net email is sponsored by: IBM Linux Tutorials
> Free Linux tutorial presented by Daniel Robbins, President and CEO of
> GenToo technologies. Learn everything from fundamentals to system
> User-mode-linux-user mailing list