[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Condor-users] How many submit nodes in your pool ?
- Date: Tue, 6 Jul 2004 18:32:08 +0100 (BST)
- From: Bruce Beckles <mbb10@xxxxxxxxx>
- Subject: Re: [Condor-users] How many submit nodes in your pool ?
Hi Nick -
Thanks for that. I was aware of both those points, and wasn't
trying to suggest having just one submit host, so much as asking whether
people routinely restrict the number of submit hosts, or whether the norm
is to have all (or most) execute hosts being submit hosts as well. I'm
sure you appreciate the flip side of having more submit hosts is that it
makes the security model more complicated, and can also make using the
pool more complicated (e.g. no shared filesystem and users submit from
multiple machines, then they need to remember where they submitted a
particular job from), and presumably may also make
troubleshhoting more complex.
I'm just trying to get a feel for how other people are negotiating
On Sun, 4 Jul 2004, Nick LeRoy wrote:
> As a Condor developer, I can give you some input. In general, more submit
> nodes is better than fewer. Why, you ask? For running a lot of simultaneous
> jobs, the schedd and shadow processes can become a significant load on the
> submit machine. For sufficiently "beafy" machines, this number can be pretty
> large (say 1k), but it can still be a bottleneck.
> Another advantage to having multiple submit nodes is that if your single
> submit node goes down, no jobs get run, but with multiple submit nodes, only
> the jobs from this dead machine stop.
> Just some food for thought.
University of Cambridge Computing Service.