Don't forget to look hard at the disk array, where's the bottleneck?
Could more spindles benefit you more than moving to Fibre Channel?
Under some conditions U320 SCSI is faster than Fibre Channel (unless you
have already gone to 2GB FC, then I'd have to re-do the math....).
Often you bottleneck is in the seek times, then the biggest thing you
can do is spread the load around to more spindles.
I do like the Squid idea.
Kevin
Magnus wrote:
On 3/4/06, Matthew Lavigne <[EMAIL PROTECTED]> wrote:
All the switches that I have are GigE, all the system GigE, HW RAID on
Ultra 160 Enclosures. I am trying to push the install in less time then
that, if I can by making the bottleneck the local CPU/Mem. Plan is to
go to U320 or FC enclosure. Working on benching that now to try to
determine which way to go.
Front-end your kickstart server with squid, and load it up with RAM. Tune
squid to cache large objects. You'll likely need to mess with kernel tuning
on the squid box (things like SHMMAX) so that squid can occupy, say, 2GB of
RAM. The most commonly requested packages will get pulled from memory.
No matter how fast your disk is, pulling from memory is faster. :-) If
you're doing a lot of kickstarts, it might be worth it to load a box up with
4-8GB of RAM.
Also look into tuning your gigabit network for jumbo frames, which will
drastically improve network throughput.
At the last place that I was at where I did a lot of kickstarts at a time,
my kickstart server had four GigE adapters that were bonded. Combine that
with a lot of RAM (remember, you actually have to tune your kickstart server
to USE the RAM), and I could kickstart a bunch of machines at once and not
bottleneck badly at all.
--
TriLUG mailing list : http://www.trilug.org/mailman/listinfo/trilug
TriLUG Organizational FAQ : http://trilug.org/faq/
TriLUG Member Services FAQ : http://members.trilug.org/services_faq/
