[Bioclusters] Advice:Several Linux Clusters and FLEXLM

Chris Dagdigian bioclusters@bioinformatics.org
Thu, 30 Sep 2004 12:39:15 -0400

{ my $.02 }

I don't see this as being any sort of problem...all of your clusters can 
talk to the flexlm server if necessary to learn what is available...

All of the flexlm integration work I've done on LSF and Grid Engine 
clusters has involved local cluster-side scripts or daemons that 
regularly poll one or more of the central, enterprise FlexLM servers.

The scripts query the server for the status and availability of each 
particlar application token they care about. This information is then 
cached/incorporated into the cluster scheduler while being regularly 

The approach is not 100% bulletproof so you typically need epilog or job 
termination scrips that are capable of automatically finding the jobs 
that died due to license-not-available reasons caused by the lightweight 
and detached nature of the polling mechanism.

If your licenses are cheap and/or under-subscribed (never found this 
yet!) you can cheat and avoid a few of the race conditions by fooling 
your cluster into thinking there are 1-2 tokens less then are really 
available. That gives you a bit of a buffer if some other cluster takes 
some free tokens between polling updates and job dispatch.

As long as all your clusters have at least one node that can poll the 
FlexLM server(s) this should work.

You should be able to follow the information you found online and just 
replicate it on each of your clusters.

If you want to get fancy you can write a single license polling daemon 
that regularly sends information in some sort of parsable form to each 
of the local clusters. Your local license checking scripts then have 
much less work to do and don't have to talk on the network.


Daniel.G.Roberts@aventis.com wrote:

> Hello All
> Anyone have any experience using FlexLM servers with several different linux clusters?
> In essence, I have three separate linux clusters which need to communicate with a flexlm server.
> I can find several examples of one cluster communicating with a flexlm server, but what about more than one cluster?
> Thanks for any direction!
> Sincerely
> Dan

Chris Dagdigian, <dag@sonsorol.org>
BioTeam  - Independent life science IT & informatics consulting
Office: 617-665-6088, Mobile: 617-877-5498, Fax: 425-699-0193
PGP KeyID: 83D4310E iChat/AIM: bioteamdag  Web: http://bioteam.net