Message boards :
ATLAS application :
BOINC downloads only 2 ATLAS tasks
Message board moderation
Author | Message |
---|---|
Send message Joined: 18 Dec 15 Posts: 1815 Credit: 118,667,202 RAC: 39,431 |
With my 12 core processor (6+6HT), 32GB RAM, I'd like to process at least three 2-core ATLAS tasks conurrently. Some time ago, it was no problem at all to download a sufficient number of tasks. Now, however, BOINC downloads only 2 ATLAS tasks, regardless whether in the web settings I put 3, 4, 5, ... or even an unlimited number of tasks. It says "no tasks available" (although the Project Status page permanently shows a few thousand unsent tasks). Which means that I can run only 2 ATLAS tasks concurrently. Any idea what's going wrong? |
Send message Joined: 27 Sep 08 Posts: 847 Credit: 692,006,372 RAC: 113,743 |
I can get 3, with no limit setting. |
Send message Joined: 15 Nov 14 Posts: 602 Credit: 24,371,321 RAC: 0 |
I am only getting 2 at a time also, regardless of the setting. Maybe it is the priority they are putting on Sixtrack? I would do more ATLAS if they will send it. |
Send message Joined: 18 Dec 15 Posts: 1815 Credit: 118,667,202 RAC: 39,431 |
So maybe there is some policy behind it; if so, it would be nice if they told us - just so that we know. |
Send message Joined: 28 Sep 04 Posts: 728 Credit: 49,143,148 RAC: 29,797 |
I think that this setting is not working the way it should. Currently I'm running on one host with 3 cores and this gets 3 tasks and an other host with 4 cores and this gets 4 tasks. They are both set up for "No Limit". A while back I was running third host with 1 core and it got just 1 task at a time. All hosts had 0.4+0.2 days cache. |
Send message Joined: 15 Nov 14 Posts: 602 Credit: 24,371,321 RAC: 0 |
I think that this setting is not working the way it should. Currently I'm running on one host with 3 cores and this gets 3 tasks and an other host with 4 cores and this gets 4 tasks. They are both set up for "No Limit". A while back I was running third host with 1 core and it got just 1 task at a time. All hosts had 0.4+0.2 days cache. That could be it. I am set for 2 cores, and am getting 2 tasks. It looks like they got them a bit mixed up. |
Send message Joined: 18 Dec 15 Posts: 1815 Credit: 118,667,202 RAC: 39,431 |
I think that this setting is not working the way it should. That could be it. yes, I agree, it seems like. So I am wondering when this will be repaired. |
Send message Joined: 18 Dec 15 Posts: 1815 Credit: 118,667,202 RAC: 39,431 |
I think that this setting is not working the way it should. Currently I'm running on one host with 3 cores and this gets 3 tasks and an other host with 4 cores and this gets 4 tasks. They are both set up for "No Limit". A while back I was running third host with 1 core and it got just 1 task That could be it. I am set for 2 cores, and am getting 2 tasks. It looks like they got them a bit mixed up.I guess we now can confirm that there is this mix-up in the settings. Because when I now changed the setting of number of cores from 2 to 3, I could download 3 ATLAS tasks. So let's hope that someone back at CERN will get this error straightened out. |
Send message Joined: 27 Sep 08 Posts: 847 Credit: 692,006,372 RAC: 113,743 |
I asked them to take a look. |
Send message Joined: 15 Nov 14 Posts: 602 Credit: 24,371,321 RAC: 0 |
I asked them to take a look. Thanks. This project actually has remarkably good support as compared to some that I could name. |
Send message Joined: 15 Nov 14 Posts: 602 Credit: 24,371,321 RAC: 0 |
There is a way to deal with this. Just set "Max # CPUs" to 8, so that it will download 8 work units. Then, use an app_config.xml to set the number of cores that you actually want to run. I usually run on 2 cores, for maximum efficiency. (This includes a max_concurrent, which is optional.) <app_config> <app> <name>ATLAS</name> <max_concurrent>4</max_concurrent> </app> <app_version> <app_name>ATLAS</app_name> <plan_class>native_mt</plan_class> <avg_ncpus>2</avg_ncpus> </app_version> </app_config> I hope they are working on a more permanent fix, with an increased number of work units. |
Send message Joined: 18 Dec 15 Posts: 1815 Credit: 118,667,202 RAC: 39,431 |
There is a way to deal with this. Just set "Max # CPUs" to 8, so that it will download 8 work units.Jim, that's an interesting approach. And it really works? Why do I ask? When doing this, each downloaded task in the BOINC Manager is shown as "Ready to start (8 CPUs)" - and these 8 CPU-tasks work as 2-core tasks (as set in the app_config)? |
Send message Joined: 15 Jun 08 Posts: 2534 Credit: 254,130,739 RAC: 54,031 |
There is a way to deal with this. Just set "Max # CPUs" to 8, so that it will download 8 work units.Jim, that's an interesting approach. And it really works? Are you aware that Jim1348 is running ATLAS (native) on his linux boxes? On a windows box you'll have to set "<plan_class>vbox64_mt_mcore_atlas</plan_class>" instead. You may check if this is set correctly. In addition you may add the line: <cmdline>--nthreads 2 --memory_size_mb xxxx</cmdline> with xxxx=4400 (according to the project's RAM formula) xxxx=5000 (safe setting if you have enough resources) |
Send message Joined: 15 Nov 14 Posts: 602 Credit: 24,371,321 RAC: 0 |
Are you aware that Jim1348 is running ATLAS (native) on his linux boxes? Thanks for the clarification. I tend to forget about Windows on LHC, though I use it on other projects. Erich, yes it works, at least in my setup. Good luck. |
Send message Joined: 18 Dec 15 Posts: 1815 Credit: 118,667,202 RAC: 39,431 |
Jim wrote: There is a way to deal with this. Just set "Max # CPUs" to 8, so that it will download 8 work units.Jim, your procedure for downloading a higher number aof ATLAS tasks works fine also here. However, I can crunch only up to three 1-core ATLAS tasks concurrently, regardless of what my settings in the app_config are. If I set it to "4" (as in your example above), still only 3 tasks are being processed. With 32GB RAM (and 6+6HT CPU cores), processing of 4 1-core ATLAS tasks concurrently should not be a problem, should it? BTW: on my system, experience shows that highest efficiency is with 1-core tasks (I had tried it also with 2-core and 3-core). |
Send message Joined: 15 Nov 14 Posts: 602 Credit: 24,371,321 RAC: 0 |
However, I can crunch only up to three 1-core ATLAS tasks concurrently, regardless of what my settings in the app_config are. If I set it to "4" (as in your example above), still only 3 tasks are being processed. Erich, I usually run 2-core tasks (three at a time). But yesterday I tried the 1-core tasks, and was able to crunch six at a time (I did not try seven). So it should be possible. I used the same app_config that I posted above, except with <avg_ncpus>1</avg_ncpus> By the way, I can now download only seven ATLAS tasks at a time rather than eight. As I recall, the first time I tried I got eight, but the first one quickly errored out, and it has been only seven since. There is something wrong with the LHC server settings anyway, so who knows how it will work in any given case. But if you want to try something else, it has been suggested to me by another very knowledgeable user (who I think likes to remain anonymous) that this is a better one, and I will try it later today: 1-core-setup <app_config> <app> <name>ATLAS</name> <max_concurrent>2</max_concurrent> # replace "2" with your favorite number <report_results_immediately/> </app> <app_version> <app_name>ATLAS</app_name> <plan_class>native_mt</plan_class> <avg_ncpus>1.0</avg_ncpus> <cmdline>--nthreads 1 --memory_size_mb 3600</cmdline> # memory_size_mb can be lower than with vbox </app_version> <project_max_concurrent>2</project_max_concurrent> # replace "2" with your favorite number <report_results_immediately/> </app_config> 2-core-setup <app_config> <app> <name>ATLAS</name> <max_concurrent>2</max_concurrent> # replace "2" with your favorite number <report_results_immediately/> </app> <app_version> <app_name>ATLAS</app_name> <plan_class>native_mt</plan_class> <avg_ncpus>2.0</avg_ncpus> <cmdline>--nthreads 2 --memory_size_mb 3600</cmdline> # memory_size_mb can be lower than with vbox </app_version> <project_max_concurrent>2</project_max_concurrent> # replace "2" with your favorite number <report_results_immediately/> </app_config> Have fun. PS - My LHC settings are now: Max # jobs No limit Max # CPUs No limit for whatever difference that makes. |
Send message Joined: 15 Jun 08 Posts: 2534 Credit: 254,130,739 RAC: 54,031 |
What I really miss is a comment from the experts at CERN. Is there anybody who takes a look at the server setup? |
Send message Joined: 18 Dec 15 Posts: 1815 Credit: 118,667,202 RAC: 39,431 |
But if you want to try something else...thanks, Jim, for the suggestion. What I am wondering, though, is the low amount of memory: 3600MB (for both 1-core and 2-core). What I have observed for long time is that - at least here - console 3 shows memory usage of up to 5900MB (so, to be on the save side, my memory setting in the app_config is 7000MB). |
Send message Joined: 15 Jun 08 Posts: 2534 Credit: 254,130,739 RAC: 54,031 |
But if you want to try something else...thanks, Jim, for the suggestion. What I am wondering, though, is the low amount of memory: 3600MB (for both 1-core and 2-core). That's exactly the pitfall. Jim runs ATLAS (native) on linux. Erich runs ATLAS (vbox...) on windows. The latter usually needs more RAM as the VM has to set up an internal CVMFS cache. |
Send message Joined: 15 Nov 14 Posts: 602 Credit: 24,371,321 RAC: 0 |
Jim runs ATLAS (native) on linux. I keep forgetting. You need more than one mind on this problem. |
©2024 CERN