[Gate-users] filemerger...

[Park MinJae]

Dear Jan,

Even though ROOT recovered some keys...there is no problem with processing
root files and there are no missing data...

Now I figure out what the problem is...
The problem is ROOT version...
Filemerger works very good with root 5.14...even without fastMerge...
Filemerger with root 5.12 works partially...such as with fastMerge option on
small root files...

I think, latest_event_id is not related with filemerger or gate
simulation...

root files from benchmark do not have any entry in latest_event_id...
but they can be merged...

My simulation data which are over 600M can be merged well...
They don't have any entity in latest_event_id...

Ah...empty latest_event_id may be related with error on ASCII option...
but I changed the job splitter source to work around it...

Special thanks to Samuel Burg and Jan De Beenhouwer...

MJ

//=================================================
//  Park MinJae
//  SNU BME Ph.D. Candidate
//
//  DialTo:+82-17-711-7809
//  MailTo:MJ at ToTi.us
//
//  Make the Dream True...
//================================================= 
 

-----Original Message-----
From: Jan De Beenhouwer [mailto:jan.debeenhouwer at ugent.be] 
Sent: Wednesday, October 24, 2007 4:13 PM
To: Park MinJae
Subject: RE: RE : RE: [Gate-users] filemerger...

On Wed, 2007-10-24 at 13:06 +0900, Park MinJae wrote:
> Dear Jan,
> 
> I did benchSPECT with jobsplitter...
> filemerger works for it...with fastMerge option on root 5.12...
> Without fastMerge option...there is same error except a message...
> following is shell output...
> ----------------------------------------
> turbo978 at MJ-GATE:~/gate_v3.1.2/benchmarks/benchmarkSPECT$
> ../../Cluster_Tools/filemerger/gjm -fastMerge SPECTbench.mac 
> turbo978 at MJ-GATE:~/gate_v3.1.2/benchmarks/benchmarkSPECT$
> ../../Cluster_Tools/filemerger/gjm SPECTbench.mac
> Combining:
> /home/turbo978/gate_v3.1.2/benchmarks/benchmarkSPECT/benchSPECT1.root
> /home/turbo978/gate_v3.1.2/benchmarks/benchmarkSPECT/benchSPECT2.root 
> -> benchSPECT.root
> 
>  *** Break *** segmentation violation
>  Generating stack trace...
>  0xb6a85309 in __libc_free + 0x49 from /lib/tls/i686/cmov/libc.so.6
>  0xb6c268b1 in operator delete(void*) + 0x21 from 
> /usr/lib/libstdc++.so.6
>  0x0804a3c5 in <unknown> from ../../Cluster_Tools/filemerger/gjm
>  0xb6a36ea2 in __libc_start_main + 0xd2 from 
> /lib/tls/i686/cmov/libc.so.6
>  0x08049db1 in __gxx_personality_v0 + 0x6d from 
> ../../Cluster_Tools/filemerger/gjm
> Aborted
> turbo978 at MJ-GATE:~/gate_v3.1.2/benchmarks/benchmarkSPECT$
> ----------------------------------------
> The merged root file with fastMerge option is able to read in 
> TBrowser...and the result is find...[600s] but root had to recover 13 
> keys...
> 


If root has to recover some keys, then something is wrong with the root
files and you will most likely be missing a lot of data.

> I dont know how to accept this...
> 

you can't


> I guess there is some problem with output option...
> 


Is there an entry in the latest_event_id histogram for the separate root
files of the benchmark before merging? If not, then the problem lies with
the way Gate is set up and compiled.

regards,


--
MScE Jan De Beenhouwer
Ghent University Hospital
IBITECH - MEDISIP
De Pintelaan 185
Block B - 5th floor
B-9000 Ghent
Tel. +32-9-240.43.26
Fax: +32-9-240.33.90
E-mail: <Jan.DeBeenhouwer at UGent.be>