Re: [ROOT] Slow down in reading TTree

From: Victor Perevoztchikov (perev@bnl.gov)
Date: Thu May 15 2003 - 01:20:57 MEST 

Hi Rene,
my apology, it was a false alarm.

New ROOT One Branch    Real: 0.13 CPU: 0.13
Old   ROOT One Branch    Real: 0.91 CPU: 0.90

New  ROOT  All branches Real: 5.65 CPU: 2.50
Old    ROOT All branches  Real: 5.04 CPU: 2.47

So you see:
1.  new ROOT for all branches apprximately the same as old one
2. new  ROOT for one branch about 10 times faster (as expected, it was a bug
in old ROOT fixed now)
3. Old   ROOT for one branch 2.5 times (only) faster then for all branches
(see the bug above)
4. New ROOT for one branch 20 times             faster then for all branches

Now why I send so bad result for new ROOT?

I looked into code. The reading of dst was overcomplicated
and inefficient. The author attempted to do by hands the job of ROOT.
His optimisation clashed with ROOT optimisation, as a result the same data
was  read a lot of times. The slowdown due to this logic about hundred
times.
And by chance this logic in new ROOT was even slower.

So sorry for false alarm again.

Victor





Victor M. Perevoztchikov   perev@bnl.gov
Brookhaven National Laboratory MS 510A PO Box 5000 Upton NY 11973-5000
tel office : 631-344-7894; fax 631-344-4206;

----- Original Message -----
From: "Rene Brun" <Rene.Brun@cern.ch>
To: "Victor Perevoztchikov" <perev@bnl.gov>
Cc: <roottalk@cern.ch>; "Jerome Lauret" <jeromel@bnl.gov>
Sent: Wednesday, May 14, 2003 12:04 PM
Subject: Re: [ROOT] Slow down in reading TTree


> Hi Victor,
>
> Please provide more information!
> Do you see this slow down only for a few branches or for any branches, all
> branches?
> Could you activate gDebug=2 when reading with 3.03 and 3.05?
> Could you send the output of
> file->ShowStreamerInfo();
>
> Rene Brun
>
> > Victor Perevoztchikov wrote:
> >
> > Dear rooter,
> > during transition from ROOT 3.03.09 ==> 3.05.04 in STAR experiment
> > we found that reading of TTree is 2.5 times slower in new version.
> > Timong
> > 3.03.09: Real: 38.42 CPU: 36.93
> > 3.05.04  Real: 93.94 CPU: 93.21
> > DST was produced by ROOT 3.03.09
> > Reading a few branches from big TTree
> >
> >
> > Any ideas why?
> > May be old format is still supported but slowly?
> >
> >
> > Victor
> >
> >
> > Victor M. Perevoztchikov   perev@bnl.gov
> > Brookhaven National Laboratory MS 510A PO Box 5000 Upton NY 11973-5000
> > tel office : 631-344-7894; fax 631-344-4206;
>

This archive was generated by hypermail 2b29 : Thu Jan 01 2004 - 17:50:11 MET