[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Condor-users] How to tweak the NEGOTIATOR_CYCLE_DELAY
- Date: Mon, 16 Jul 2007 09:57:49 +0100
- From: "Matt Hope" <matthew.hope@xxxxxxxxx>
- Subject: Re: [Condor-users] How to tweak the NEGOTIATOR_CYCLE_DELAY
On 7/16/07, Vincent Tollu <vincent.tollu@xxxxxxxxx> wrote:
Pretty new in grid processing, I'm currently making some tests with Condor.
Having a bunch of really small jobs (a few seconds only) to run, I'm facing
what seems a uncompressible delay of 20 seconds for my jobs.
You are trying to tweak something not designed for the task you are
performing. You can improve things with tweaks but you are going to be
wasting vast amounts of throughput no matter what.
There is a 'wrapper' around condor by technion which aims to handle
the many small jobs (essentially by batching them up behind your back
and trying to spot reuse of the same data). You would do much better
to try to make your jobs much more chunky (why retransmit the
executable every time if it is always the same one)
Lurking around, I concluded I might have to tweak the NEGOTIATOR_CYCLE_DELAY
You first need to use the NEGOTIATOR_INTERVAL setting to reduce the
interval. Then you can tweak the cycle delay (a forced minimum wait
between cycles to allow the state changes form the last negotiation
actions to percolate to the collector - unless you really know what
you are doing don't mess with it.
You also need to look at your claim timeouts and queue lengths but
basically you should first of all try to increase the length of your
jobs. This is the easiest and most stable way of increasing