Re: [ROOT] PROOF, TDSet and a large file

From: Fons Rademakers (
Date: Tue Mar 16 2004 - 14:32:57 MET

Hi Axel,

   all depends how large this single file is and on the time spend
processing an event. In principle PROOF works fine with a single file
and the different slaves will access the same file at different event
offsets. Note that PROOF will always access a file via the rootd
protocol for performance reason (generally better than nfs).

Currently friend chains or not handled yet. It is the next thing on our
todo list.

Rewriting your code to use a TSelector will always pay off since you
will be able to profit from PROOF.

Cheers, Fons.

On Tue, 2004-03-16 at 11:48, Axel Naumann wrote:
> Hi,
> I posted a question to the board, no answer, so I'll try it the old way:
> I have a little proof cluster, all running cvs root on linux, all 
> sharing the disks for code and data via NFS. I'd like to know whether 
> running on a single file is accelerated by a proof cluster, e.g. by each 
> of the slaves skipping (number of slaves) events, starting with a given 
> offset. Would this be easy to implement? If the number of active slaves 
> is coupled to the number of files (I somewhat remember that): suppose I 
> have a TDSet'ed chain, and I add a single friend tree (with as many 
> entries as the whole chain) to it - is that handled somehow, or do I end 
> up with a single slave running the whole job?
> I know I could just try all of that, but before trying I'd need to spend 
> some (non negligible) time re-writing code, and I'd like to know whether 
> that pays off before investing that time...
> If all that is documented then I just didn't find it. Pointing me to it 
> would be sufficient ;-)
> Axel.
Org:    CERN, European Laboratory for Particle Physics.
Mail:   1211 Geneve 23, Switzerland
E-Mail:              Phone: +41 22 7679248
WWW:      Fax:   +41 22 7679480

This archive was generated by hypermail 2b29 : Sun Jan 02 2005 - 05:50:06 MET