Message boards : CMS Application : New Version 60.64
Message board moderation
| Author | Message |
|---|---|
 Laurence CERN Send message Joined: 12 Sep 14 Posts: 1150 Credit: 342,270 RAC: 2  |
This new version should have the same CVMFS and proxy configuration as on LHC@home. |
Magic Quantum Mechanic Send message Joined: 8 Apr 15 Posts: 830 Credit: 15,276,826 RAC: 16,920  
|
I knew there must be a reason I get awaken at 3am And my high speed won't start for 48 hours so I can get this taken care of before it is time to run some. Good time to start the d/l so now I can go back to sleep  |
|
Send message Joined: 13 Feb 15 Posts: 1252 Credit: 995,382 RAC: 50 |
The information that should have been shown, when using ALT-key's is still not displayed in this version. ALT-F1 Startup console is shown ALT-F2 shows only the dummy message (job output - event processing) ALT-F3 "top" is shown correctly ALT-F4 shows only the dummy message (output of job wrapper) ALT-F5 shows only the dummy message (error messages) ALT-F6 login screen is shown |
|
Laurence CERN Send message Joined: 12 Sep 14 Posts: 1150 Credit: 342,270 RAC: 2 |
The consoles should now be working. The information that should have been shown, when using ALT-key's is still not displayed in this version. |
 Send message Joined: 28 Jul 16 Posts: 527 Credit: 400,710 RAC: 0 |
The consoles should now be working. Yes, they are working. |
|
Magic Quantum Mechanic Send message Joined: 8 Apr 15 Posts: 830 Credit: 15,276,826 RAC: 16,920
|
Yes , nice seeing the log page again. |
|
Send message Joined: 28 Jul 16 Posts: 527 Credit: 400,710 RAC: 0 |
The consoles are now working on -dev, but on -prod they don't work any more. ??-| |
|
Send message Joined: 17 Aug 15 Posts: 17 Credit: 228,358 RAC: 0
|
I am running four cores on one 60.64, but BoincTasks shows that only one core is in use. (Ryzen 3900X, Ubuntu 20.04.4, BOINC 7.16.6, VBox 6.1.34) And the estimated time shows as 18 hours total, with 1 hour completed. It appears that mt is not working properly. |
|
Send message Joined: 28 Jul 16 Posts: 527 Credit: 400,710 RAC: 0 |
CMS-dev and CMS-prod get their subtasks from the same backend system pool. This pool is configured to generate 1 subtask per VM. Hence, you will just get a slightly higher CPU time than runtime and idle CPUs. |
|
Send message Joined: 17 Aug 15 Posts: 17 Credit: 228,358 RAC: 0
|
I suppose this will be fixed in production, so that you get the benefit of the additional cores? I can just wait until then. Thanks. |
|
Send message Joined: 28 Jul 16 Posts: 527 Credit: 400,710 RAC: 0 |
I suppose this will be fixed in production No. At least not not in the near future. Unlike ATLAS which runs real threads CMS subtasks are singlecore. The multicore VM here was to test (long ago) whether they can be run concurrently. As far as Ivan wrote this is not very efficient for more than 2 subtasks. In addition the console monitoring can only show 1 of them. We would run the others 'blind' and would have to explain every day why the VM runs 100 % of n cores although the consoles show 'the' subtask is finished. |
|
Send message Joined: 17 Aug 15 Posts: 17 Credit: 228,358 RAC: 0
|
That is all very interesting, if a little strange. I would hope that in production, they limit the CMS to a single core in that case, so that you can select more than one core for native ATLAS. |
|
Send message Joined: 28 Jul 16 Posts: 527 Credit: 400,710 RAC: 0 |
Unlike ATLAS CMS requires the #cores to be set at creation time (before tasks are sent to a client). For quite a while a 4-core configuration seems to be a good compromise between performance and total runtime. Changing the #cores or creating batches with different #core settings (which has been tested) just causes a lot of weird trouble on the BOINC client side as well as on other task levels like glidein. Hence this should be avoided. Since the CMS queue feeds tasks from dev as well as from prod the #cores is the same for both projects. |
©2025 CERN