Winnie Lacesso wrote:
> David,
>
> Thanks very much for the info.
> Sounds like you're where I want to be but I'm just starting out.
>
> That's all bridged networking right?
> For the ones that are similar (cores/ram/disk) did you make one VM, install
> OS & then make clones, or all separate VMs & OS installs?
>
> Your installhost runs DHCP server right? I'm trying to picture
> the networking. Is there an internal virtual router?
>
> How is the load on vhost1 & vhost2, are they ever approaching high/loaded?
>
> Thank you**2 for info.
>
We have a separate node "master" which runs the DHCP, DNS, and NFS for
the subnet, which is a real machine, and then the next-server line
points to the installhost
We spent quite a bit of time in making the nodes (and indeed frontends)
automatically install completely unattended. We can re-kickstart a
worker node in around 20-30 minutes (of which only a couple require our
attention), so we've just installed everything from scratch.
Kickstart brings up a base install, CFengine installs and configures the
nodes to our requirements.
All networking is bridged, and the cluster (frontends and nodes) sits on
a publicly routable /24.
Feel free to have a poke around our ganglia
(http://ganglia.dur.scotgrid.ac.uk/ganglia) Installhost is currently
causing the vhost2 to get loaded, but thats only because I'm prepping
for the SL5 upgrade. Overall performance of the actual site appears
reasonable.
The only machine we virtualised but had second thoughts about was our
DPM headnode, but we don't know how the site would perform with a modern
real DPM headnode either (as the headnode it replaced was very old)
--
David Ambrose-Griffith - [log in to unmask]
Assistant Systems Programmer,
IPPP, Department of Physics, Durham University,
Science Laboratories, South Road, Durham, DH1 3LE
Direct Dial: +44 (0)191 3343704
Office: +44 (0)191 334 3811
|