I’m doing an analysis with proof, it works and it is very fast, it takes 5 minutes. I’m working from office, and our proof cluster is very near to my office pc. What I do is to log into the proof cluster where the data are stored and run the session with TProof::Open(“localhost”);
I tried to do the same thing from CERN, and nothing changed. But when I tried from home the same analysis becomes very very slow, it takes more than one hour! If I look at the net traffic plot on my home pc I see that it is totally full! Why? The only things proof has to communicate is the graphical indicator.
I tried to disable X, the traffic becomes smaller, but the analysis is still too slow.
Remote X can indeed be a pain. However, running in batch mode (root -b) should not be so dramatic.
What do you mean exactly by disabling X? Can you quantify the effect?
Ok, then try by running root in batch mode (root -b): you should get a text progress bar from PROOF (like the one for copying files from scp, for example) instead of the progress window. The performance should be OK.
[quote=“ganis”][quote]
I mean that I logged into the cluster with ssh without -X (-Y)
[/quote]
Ok, then try by running root in batch mode (root -b): you should get a text progress bar from PROOF (like the one for copying files from scp, for example) instead of the progress window. The performance should be OK.
G. Ganis[/quote]
This is what I already get. ROOT says DISPLAY not set blabla and I got a textual progress bar, the DSL band is not full, but the perfomances are not good, al least 1/2 compared from the analysis from office
Unfortunately I am not able to reproduce the problem.
If I run test/stressProof (which does not use any X) on my office desktop from home via ADSL and locally I get exactly the same processing times.
Can you try this?
$ cd $ROOTSYS/test
$ make stressProof
$ ./stressProof
Just to see if for you the problem is already at the level of these standard tests ...
[quote=“ganis”]Unfortunately I am not able to reproduce the problem.
If I run test/stressProof (which does not use any X) on my office desktop from home via ADSL and locally I get exactly the same processing times.
Can you try this?
$ cd $ROOTSYS/test
$ make stressProof
$ ./stressProof
Just to see if for you the problem is already at the level of these standard tests ...
[quote=“ganis”][quote]
Yes, with this test the behaviour is very similar
[/quote]
Which means? You see the performance difference or not?
G. Ganis[/quote]
Yes, I see the same performance. Now I retried my selector from home and I get the same results as from office. It’s strange, maybe there were some problems on our cluster last time.