Re: Chain or tree?

From: Rene Brun <Rene.Brun_at_cern.ch>
Date: Thu, 25 Oct 2007 14:59:30 +0200


Chiara,

Let me put this way ::)
-It is better to have one single Tree of 1 GByte than a TChain of 1000 files with a 1 Mbyte Tree.
-It is better to have one TChain with 1000 files with a 1GByte Tree in each than one single Tree of 1 TeraByte.

Larger the Tree, larger the internal tables to address the Tree baskets. Anyhow the overhead indiced by a TChain should be very small (opening/closing files),
TChain has many advantages in case you want to parallelize the processing. The individual files
can be on different nodes. You do not need a gigantic Tree on one single node that will generate
I/O bottlenecks, etc.

Rene Brun

Chiara Zampolli wrote:
> Dear all,
>
> I was wondering whether there is a difference in performance
> (memory, mainly), in case one chains some trees written on files using
> a TChain, or building a TTree with CopyEntries. I have tried to see
> what happens using gSystem->GetMemInfo(), and also with "top", but it
> seems as if there's no difference.... Am I wrong? BTW, which is the
> "best" solution?
>
> Thanks in advance. Cheers,
> Chiara
>
Received on Thu Oct 25 2007 - 14:59:28 CEST

This archive was generated by hypermail 2.2.0 : Thu Oct 25 2007 - 17:50:02 CEST