Configuring a small PROOF cluster

Dear All,

I’m trying to configure a small PROOF cluster (currently containing two machines), but I’m stuck. Maybe someone here will be able to help. (With Gerri still on holiday. :wink:)

I have two PCs in my office, one quad-core, pretty high-powered one, and a pretty old one. I’m just using the second one as a “proof of concept”. (No pun intended… :stuck_out_tongue:) I’ve set up single machine “PROOF clusters” before, because I wanted to be able to send jobs to a machine from multiple other machines. That always worked well. Now I’d like to achieve the following setup:

  • The bigger machine should act as a master, and it should also run 4 worker threads.
  • The smaller machine should run one worker thread.
  • The bigger machine has a big external disk that should be visible to all the nodes in the same way. (Only read access.)
  • Both machines should have some local storage for the output of using TProofOutputFile.

I’ve created a configuration which you can find under

/afs/cern.ch/user/k/krasznaa/public/nyu_cluster/

PROOF actually seems to work, but when my application tries to collect the output ntuples from the worker nodes, it complains that it can’t access

root://peppe26.cern.ch//home/proof/kras … 9beef.root

(peppe26.cern.ch is the “smaller” machine.)

So the smaller machine is definitely not acting as a file server toward the master. I’m trying to find the solution in the online PROOF documentation, but these configuration issues are not described all that well I’m afraid. (At least I always make my configuration in a trial-and-error fashion…)

You can find the full logfile of this under

/afs/cern.ch/user/k/krasznaa/public/proof/FirstCycle.log

Any help would be much appreciated.

Cheers,
Attila

For the record, attached is a configuration file that solved Attila’s problem.

Gerri
xrootd.cf.txt (1.23 KB)

Hello Gerri & Attila

I have the same problem: when trying to merge, the master doesn’t manage to retrieve the files located on the slave nodes.

example: the master is ccapl0001 , it is trying to fecth a file on ccapl0003

TProofOutputFile::AddFile: error from TFileMerger::AddFile(root://ccapl0003.in2p3.fr//scratch/proo … 6//Pt.root)

I have tried the config file posted by Gerri, trying every option line by line, but it doesn’t help.
What is the relevant line for this special case ?

Also, there is one thing I don’t understand : what is supposed to do the /pool0 directory ?
It is relevant for this matter ?

Many thanks !
Renaud