Message boards : Theory Application : Theory native - singlecore vs. multicore
Message board moderation

To post messages, you must log in.

AuthorMessage
computezrmle
Volunteer moderator
Project tester
Volunteer developer
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 28 Jul 16
Posts: 481
Credit: 394,720
RAC: 4
Message 5879 - Posted: 13 Feb 2019, 8:27:42 UTC

It looks like all scientific apps running under the roof of Theory Simulation are designed to run as singlecore app or at most as a chain of different singlecore apps. Even Theory (vbox) is not really a multicore app as it just provides a couple of slots each of them is running independent scientific chains.

Thus I wonder why Theory native has been started as multicore.
The tests show that multicore configurations always (ATM) run the standard singlecore scientific chain.


If we could vote, I would vote to continue Theory native as a singlecore app for the following reasons:
- easier app development
- easier helper apps, e.g. BOINC validator
- easier error handling - Which "slot" caused an error? How to calculate the credits?
- less idle cores - a single longrunner (e.g. sherpa) in an n-core setup can cause (n-1) idle cores


Other comments are welcome.
ID: 5879 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Crystal Pellet
Volunteer tester

Send message
Joined: 13 Feb 15
Posts: 1188
Credit: 853,991
RAC: 130
Message 5881 - Posted: 13 Feb 2019, 10:20:30 UTC - in response to Message 5879.  

Other comments are welcome.

Replied to a post of captainjack: https://lhcathomedev.cern.ch/lhcathome-dev/forum_thread.php?id=446&postid=5872#5872

I agree the best option is single core, but I think the user should be able to run a single core task using 2 threads.
Because one job has several processes, that speeds up a single job, but one can do lesser jobs.
ID: 5881 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Laurence
Project administrator
Project developer
Project tester
Avatar

Send message
Joined: 12 Sep 14
Posts: 1067
Credit: 334,882
RAC: 86
Message 5882 - Posted: 13 Feb 2019, 13:23:58 UTC - in response to Message 5879.  

It looks like all scientific apps running under the roof of Theory Simulation are designed to run as singlecore app or at most as a chain of different singlecore apps. Even Theory (vbox) is not really a multicore app as it just provides a couple of slots each of them is running independent scientific chains.

Thus I wonder why Theory native has been started as multicore.
The tests show that multicore configurations always (ATM) run the standard singlecore scientific chain.


If we could vote, I would vote to continue Theory native as a singlecore app for the following reasons:
- easier app development
- easier helper apps, e.g. BOINC validator
- easier error handling - Which "slot" caused an error? How to calculate the credits?
- less idle cores - a single longrunner (e.g. sherpa) in an n-core setup can cause (n-1) idle cores


Other comments are welcome.


The multicore planclass was used as it already exists for ATLAS. We can create another one for singlecore.
ID: 5882 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Magic Quantum Mechanic
Avatar

Send message
Joined: 8 Apr 15
Posts: 769
Credit: 11,873,251
RAC: 5,591
Message 5885 - Posted: 14 Feb 2019, 2:07:47 UTC - in response to Message 5881.  

Other comments are welcome.

Replied to a post of captainjack: https://lhcathomedev.cern.ch/lhcathome-dev/forum_thread.php?id=446&postid=5872#5872

I agree the best option is single core, but I think the user should be able to run a single core task using 2 threads.
Because one job has several processes, that speeds up a single job, but one can do lesser jobs.


I don't mind the singles but do prefer running X2 WU's on all my 8-cores since they have the ram needed.
ID: 5885 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
computezrmle
Volunteer moderator
Project tester
Volunteer developer
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 28 Jul 16
Posts: 481
Credit: 394,720
RAC: 4
Message 5886 - Posted: 14 Feb 2019, 6:17:31 UTC - in response to Message 5885.  

I don't mind the singles but do prefer running X2 WU's on all my 8-cores since they have the ram needed.

If your host has enough RAM to run vbox tasks then it is not an issue for native tasks.
ID: 5886 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Laurence
Project administrator
Project developer
Project tester
Avatar

Send message
Joined: 12 Sep 14
Posts: 1067
Credit: 334,882
RAC: 86
Message 5895 - Posted: 14 Feb 2019, 10:49:33 UTC - in response to Message 5881.  

Other comments are welcome.

Replied to a post of captainjack: https://lhcathomedev.cern.ch/lhcathome-dev/forum_thread.php?id=446&postid=5872#5872

I agree the best option is single core, but I think the user should be able to run a single core task using 2 threads.
Because one job has several processes, that speeds up a single job, but one can do lesser jobs.


I have created the following plan class for this app:

  <plan_class>
    <name>native_theory</name>
    <min_ncpus>1</min_ncpus>
    <max_threads>2</max_threads>
    <mem_usage_base_mb>1000</mem_usage_base_mb>
    <projected_flops_scale/>
    <nthreads_cmdline/>
  </plan_class>
ID: 5895 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Magic Quantum Mechanic
Avatar

Send message
Joined: 8 Apr 15
Posts: 769
Credit: 11,873,251
RAC: 5,591
Message 5897 - Posted: 14 Feb 2019, 11:02:13 UTC - in response to Message 5886.  

I don't mind the singles but do prefer running X2 WU's on all my 8-cores since they have the ram needed.

If your host has enough RAM to run vbox tasks then it is not an issue for native tasks.


Yes of course I know that.
It means I don't care which way we run the tasks native or not.
ID: 5897 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
computezrmle
Volunteer moderator
Project tester
Volunteer developer
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 28 Jul 16
Posts: 481
Credit: 394,720
RAC: 4
Message 5898 - Posted: 14 Feb 2019, 11:19:36 UTC - in response to Message 5895.  

I have created the following plan class for this app:

  <plan_class>
    <name>native_theory</name>
    <min_ncpus>1</min_ncpus>
    <max_threads>2</max_threads>
    <mem_usage_base_mb>1000</mem_usage_base_mb>
    <projected_flops_scale/>
    <nthreads_cmdline/>
  </plan_class>

Now v4.08 and v4.09 are listed at the app page.
I got a task for v4.08.
ID: 5898 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Laurence
Project administrator
Project developer
Project tester
Avatar

Send message
Joined: 12 Sep 14
Posts: 1067
Credit: 334,882
RAC: 86
Message 5900 - Posted: 14 Feb 2019, 12:19:32 UTC - in response to Message 5898.  

Now v4.08 and v4.09 are listed at the app page.
I got a task for v4.08.


Thanks, have now deprecated all the old versions.
ID: 5900 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote

Message boards : Theory Application : Theory native - singlecore vs. multicore


©2024 CERN