PROOF and LXBatch

I am just starting to learn PROOF from http://wiki.gsi.de/cgi-bin/view/Grid/TheParallelRootFacility
and http://wiki.gsi.de/cgi-bin/view/Grid/TheParallelRootFacility.
I would like to run PROOF from my machine networked at CERN using workers in the LXBatch system( My only experience with lxbatch is using bsub commands from Lxplus machines).

Assuming that I have a working PAR file how do I set this up? If you know of any other good sources of PRROf information, please send me the url.
Thank you to everone who has been helping me out here on Root Talk.

Is there another openly accesible cluster that I can run proof on?

Hi,

Unfortunately we do not have public PROOF-enables cluster where you can run. To give a try at CERN, perhaps the best is to do it using a few lxplus machines. There is some doc at root.cern.ch/twiki/bin/view/ROOT … stallation about setting a small cluster.

For what relates to LSF, I cannot open the pages that you give so I do not know what you could find there about running PROOF on a LSF-enabled system. That was not the original idea of PROOF, so you need some trick to do that. Below are some guidelines following an (successful) attempt that I did several years ago. I hope it still works, but there is no guarantee.

The first thing to understand is if the lxbatch machines can be contacted from outside on port 1093 (or some other port): this is not obvious, given the recent restrictions on machine interconnectivity applied at CERN.

If there is a port that you can use, then you need to submit batch jobs starting the ‘proofd’ daemons: see the attachement for an example where you have to put the ROOT distribution that you want to run and substitute 1093 with the port that you are allowed to use (if different).

Once the proofd are started and running you have to get a list of them and create a PROOF configuration file: as LXBATCH uses AFS home directories (if I am not mistaken) the best is to create the file $(YOUR_AFS_HOME)/.proof.conf (note the ‘.’ before proof.conf). The file should contain a list of fully qualified machine names defining the master and the workers: something like

master   lxbxxxx.cern.ch

worker    lxbyy01.cern.ch port=1093
worker    lxbyy02.cern.ch port=1093
worker    lxbyy03.cern.ch port=1093

where you should substitute ‘lxbxxxx’ with the name of the machine chosen as master, and the ‘lxbyy0i’ with the names of the machines chosen as workers. The best would be to have a script doing this parsing the output of the LSF command ‘blist’ (or whatever is used to get the status of your jobs).

After that you should be able to connect: start a ROOT session and open a PROOF session:

root[] p = TProof::Open("lxbxxxx.cern.ch")

If this is successfully, you can upload and enable the PAR file:

root[] p->UploadPackage("MyPar.par")
root[] p->EnablePackage("MyPar")

where ‘MyPar’ is the name of the package (and of the PAR file).

Hope it helps. Let us know if you give a try.

G. Ganis
proofd_lsf.txt (349 Bytes)

Hello Geri,

the sites you mentioned above should be public. Can you please retry to open them?

@dlayton : Thanks for reading our documentation :slight_smile:
If you have further questions to our PROOF environment, just
contact me. Currently we are moving to use PROOF over xrootd.

Cheers,
Carsten (@GSI)