Mark Weinstein mark@dotcomlive.net writes:
Eric,
Thanks for the replies. I must no have made my intentions clear. I actually do NOT want to boot from a local hard drive, but I wasn¹t sure if the network boot would be way to taxing on the network. I will be using a 100MB switched network and I didn¹t know how much bandwidth it would take to remote boot the nodes. On top of that, what about any swap space that might be needed?
Note, while I reply to people when I am in a friendly mood, many times I too busy to talk to people, so please talk to the list.
With multicast booting the scalability is fine. I actually would have booted faster if the cisco switches had not been in slow port mode delaying the inferface boot speed by 30 seconds or so.
As for swap if you are seriously using that on a compute cluster you have problems. Although a little swapping to throw unused programs out is o.k.
If your main filesystem is on NFS you probably have problems unless you have a small cluster. At the moment there is not a mature general purpose solution to the bandwidth limitation problems. But it depends a lot on your application, and your NFS servers where you see limitations, or which of the partial solutions are right for you.
Eric
On 21 Oct 2002, Eric W. Biederman wrote:
As for swap if you are seriously using that on a compute cluster you have problems. Although a little swapping to throw unused programs out is o.k.
no swap. It's that simple.
no NFS either. Just avoid it like the plague.
I really think you ought to check out http://www.clustermatic.org ...
ron
-
ebiederman＠lnxi.com -
Ronald G Minnich