[Brahms-dev-l] Re: proofserver

What do you mean " jobs at KU are not fast if I read the files on 
/brahms/data** diskes".  You can not read RCF disk from KU, can't you?  
Here is the trick to make loop faster in proof (or any other):  Load 
only variables that you use to proof.  For example, in our dst, if you 
need only some of global variables (say run number, centrality), you can 
do the following in your Init method of your TSelector class.

after a line with fChain->SetMakeClass(1)

fChain->SetBranchStatus("*",0);  // this disable all variables to be 
loaded in your class.
fChain->SetBranchStatus("R.fRunNumber",1);
fChain->SetBranchStatus("G.fCent",1);


And, comment out any other SetBranchStatus command to load variables in 
that method.


Now, obviously, you can not read any faster than the hardware IO limit 
(network, server, disks etc..), but you should be able to read files 
quite fast, especially from PANASAS disk (/brahms/data21).    If you 
think your jobs are running slow (without competing jobs), it is 
probably due to IO problem.  As I said in the other mail, some 
brahms/dataXX disks are slow.  You can easily see this by "top" 
command.  Run "top" at the machines your jobs are actually running.  If 
you see a more than a few % of IO wait, your IO is the choke point, 
which you probably can not do anything to improve.  If you see that 
problem, the best option(???) is probably not to run your program since 
it just make things worse and ask offending person(s) to stop some of 
jobs (if you can find).

Hiro

Eun-Joo Kim wrote:

>I am reading the output of geant, and it is 50M events.
>Maybe I don't need 50M events.
>Anyway it took more than 2 hours, and I don't know why.
>I think making/killing program is good idea, but just
>let you know sometimes the jobs at KU are not fast
>if I read the files on /brahms/data** diskes.
>
>Eun-Joo
>  
>


_______________________________________________
Brahms-dev-l mailing list
Brahms-dev-l@lists.bnl.gov
http://lists.bnl.gov/mailman/listinfo/brahms-dev-l