Skip to Content.
Sympa Menu

sphenix-tracking-l - Re: [Sphenix-tracking-l] TPC Clusterizer performance

sphenix-tracking-l AT lists.bnl.gov

Subject: sPHENIX tracking discussion

List archive

Chronological Thread  
  • From: Christof E Roland <cer AT mit.edu>
  • To: "Osborn, Joe" <osbornjd AT ornl.gov>
  • Cc: Hugo Pereira Da Costa via sPHENIX-tracking-l <sphenix-tracking-l AT lists.bnl.gov>
  • Subject: Re: [Sphenix-tracking-l] TPC Clusterizer performance
  • Date: Mon, 23 May 2022 13:17:33 +0000

Hi Joe, 

agreed. Organizing our storage containers better will give benefits across the board. 
For the Clusterizers its a bit worse still. The performance depends a lot on how we store the hit data. 
For reliable performance studies of the cluterizers in view of our computing resources for actual data 
taking we need a format that is as close to actual raw data as possible. 
Right now I store a vector containing objects with 3 short ints for iphi, time bin and adc.
Actual raw data is probably even more compressed.

Cheers

   Christof 

On 23. May 2022, at 13:53, Osborn, Joe <osbornjd AT ornl.gov> wrote:

Hi Christof,
 
This basically tracks with everything else we’ve seen. After switching the seeding to vector containers, the modules that run over multiple containers (e.g. the tpc/silicon track matching) run way faster. Hugo saw substantial gains too when moving the cluster container to vectors. I think making these changes will prove substantial in the end once we add everything up.
 
---------------------------
 
Joe Osborn, Ph.D.
Associate Research Scientist
Oak Ridge National Laboratory
(859)-433-8738
 
 

From: sPHENIX-tracking-l <sphenix-tracking-l-bounces AT lists.bnl.gov> on behalf of Christof E Roland via sPHENIX-tracking-l <sphenix-tracking-l AT lists.bnl.gov>
Date: Monday, May 23, 2022 at 6:16 AM
To: Hugo Pereira Da Costa via sPHENIX-tracking-l <sphenix-tracking-l AT lists.bnl.gov>
Subject: [EXTERNAL] [Sphenix-tracking-l] TPC Clusterizer performance

Hi Everybody, 

I ran a test of the tpcclusterizer performance using verctors to store the hit infformation instead of our 
usual maps base hit storage containers. The performance change is rather dramatic. 
The memory goes down from 4 to 2,4 GB/job and the execution time when processing the 
tpc sectors in sequence (i.e. turning the threading off) goes down from 1.6 sec to 0.27 sec. 
See slide attached below. 

Cheers

    Christof 

Attachment: smime.p7s
Description: S/MIME cryptographic signature




Archive powered by MHonArc 2.6.24.

Top of Page