[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Condor-users] Condor Learning Curve

On Tue, May 4, 2010 at 1:46 PM, Matt Hope <Matt.Hope@xxxxxxxxxxxxxxx> wrote:

From: condor-users-bounces@xxxxxxxxxxx [mailto:condor-users-bounces@xxxxxxxxxxx] On Behalf Of Ian Chesal
Sent: 04 May 2010 16:47
To: Condor-Users Mail List
Subject: Re: [Condor-users] Condor Learning Curve

> I've been using Condor for (gasp!) over 5 years now.

Errrk, Am I that long too? :/

I know? Where has the time gone?
> It was quick to setup and get started with.

I ended up wrapping the whole thing in c# and automating everything so it wasn't that quick to get going but it means I can alter things around it very quickly. Learning what is/isn't cheap is pretty important on this though, and there are many quirks.

In a similar style, our Condor install is wrapped in Perl (really it's an XMLRPC interface). But we opted for less flexibility because we implemented a pretty strict scheduling algorithm that isn't Condor's default fairshare algorithm. Now that we're hooked on hooks we're completely bypassing Condor's negotiator and schedd daemons. Straight job pulls from our DBs by the startds.
> Administering it was a process that's evolved over time.

:) my evolution was to get someone else to admin it.

Oh man. I have to get me one of those: a lackey! :)

> By far the biggest challenge for me is keeping up with the changes in Condor.

For us the biggest challenge is performance of the negotiation/schedd/shadow parts. This is something of a constant battle.

Matt runs bigger pools than I do. :) When 6.8.x came out I was definitely comfortable with the performance and with hooks the performance became my problem and my problem alone, not Condor's.

- Ian