On 02/28/2010 09:36 AM, Eric Shubert wrote:
Jake Vickers wrote:
On 02/27/2010 12:08 PM, MagicWISP Sales wrote:
Yes the old server was a Quad PII - 4 Intel 400Mhz processors, it
ran like a
champ with QMT, Apache, Bind, and FreeRadius. As soon as the
Apache, Bind,
and QMT were rolled into VMWare -it became a dog. I used about
every tweak
I could find on the internet to make it work better. I think the
problem is
VMWare Server 2 - it's just slow. Unfortunately I could not use Xen
or KVM
because of the old processors. From what I could research, they
would have
worked much better. I am pretty anxious to see how this runs on
another
server with dual Intel 3.Ghz processors. Unfortunately they aren't
dual
core, but we should still see some marked improvement.
I'm actually testing/using VirtualBox for things right now. I think
in the long run that running a large mailserver on any virtual
environment is a bad idea.
Keyword: large. (Please be a little more specific)
I'm not inclined to accept this premise yet. It's certain that any
host needs to be adjusted in certain ways to run efficiently as a
guest. Achieving high performance in a virtual guest will also likely
require the latest innovations in virtualization hardware
(paravirtualization). With a QMT guest, the need for
paravirtualization hardware is appearing to me to be more of a
certainty in order to reach high levels of efficiency, but I'm not
positive about this yet - just a hunch. BL, I wouldn't throw in the
towel on QMT virtualization in a "large" environment quite yet. (And
to be clear, virtualized QMTs appears to be working nicely in small to
medium environments at this point).
You are correct that "large" needs to be quantified. In the scenarios
that I've been brought into, it has been either ~200 users all using
IMAP or ~500 users using mixed. Initially it was thought to be a
NFS/RAID bottleneck and after moving to attached storage the performance
increased by 200% or so. Still was not enough to correct *all* the
issues, but it did solve 85% of them.
Other tweaks were done, all from a guest perspective. I did not have any
direct interaction with the host system (using ESX, VMware Server, and
Xen), so I cannot say for sure what tuning needed to be done on those
systems. I'd hate to think all 3 of the "biggies" had the same exact
bottlenecks, but we cannot assume anything. I'm not a big virtualization
guy - I just know systems with the same metrics work without an issue on
dedicated machines.
Of course any host is going to run faster on bare iron. The question
is, how much of a performance hit are you willing to give up to gain
the benefits of virtualization? This is not a one size fits all
situation.
Correct, but I think it's a safe "blanket" statement to say that at this
time, large* systems should be dedicated. (*large still needing to be
quantified). There is a whole other world of variables and tuning that
vitualization adds to the soup mix. Until we can identify what the
actual issues are, we need to tread carefully.
On the 200 user IMAP system, even performing a simple "du -sh *" on the
mailstore took *minutes* on some dirs, with no reasoning behind it. One
dir took 6 minutes and was 6G, and another that was 24G took 1 min or
so. There may be another guest hogging the cycles during that, which
adds a whole other level of complexity to the mix.
I heard a fellow say yesterday that with VirtualBox you take a 50%
performance hit on the guest. I don't think I'd find that acceptable
that in any case. I expect though that this is with no tuning, and
that with some tuning that hit can be reduced. I haven't tried out
VirtualBox yet, so I'm glad Jake's doing so. My impression though is
that VirtualBox is stronger (grew out of) the desktop application
arena, and is being retro-fitted for server applications. On the other
hand, VMware started with servers and grew into desktops. This is
largely why I expect that VMware Server will continue to be better
suited to server virtualization than VB. That could change with time,
but I'm not expecting it any time soon, especially with Oracle-Sun
merger.
Not sure about the 50% performance hit; I know the install base is
*much* smaller than VMware server, and seems to run a little faster on
my laptop. I only use virtualization for development and testing at this
point, so cannot say what actual metrics would be either way. I
personally think running a big mail environment on a VM (VMware, Xen,
VirtualBox, et al) does not fit any of my applications and have no
desire to try and prove myself wrong, so this would require outside
testing from the community.
I'll need to verify, but I think even the "big official" mail server
packages are not supported if run in virtual environment and probably
for reasons like this.
I can think of other reasons that I think are more likely. ;)
Other than poor performance and nightmare-ish issues to troubleshoot? ;)
---------------------------------------------------------------------------------
Qmailtoaster is sponsored by Vickers Consulting Group
(www.vickersconsulting.com)
Vickers Consulting Group offers Qmailtoaster support and installations.
If you need professional help with your setup, contact them today!
---------------------------------------------------------------------------------
Please visit qmailtoaster.com for the latest news, updates, and packages.
To unsubscribe, e-mail: qmailtoaster-list-unsubscr...@qmailtoaster.com
For additional commands, e-mail: qmailtoaster-list-h...@qmailtoaster.com
