![]() ![]() That would be for the project to implement (some have), not for BOINC. If so, you should ask them to implement some form of selection process. This system is setup to have 2 CPU Groups, each of them has 36 threads. CPU: 2 x Intel Xeon E5-2699 v3 (72 threads in total) BOINC v7.4.36 å4. Rosetta runs many different research projects in parallel, and your problem might relate to just one of them. Hello, I have noticed a problem where BOINC doesnt seem to properly schedule threads on systems that have multiple CPU groups. Notice also that both of the Rosetta tasks in that display use far less that 4 GB ram, but they do use different amounts: one is a dt_ task, the other is a niv_ task. Ive changed to 12 in ccconfig.xml and verified only 12 tasks are running, but they still seem to be hitting all cores.That's certainly how the Windows tools display it: (your words, from #4096 (comment)) My computer science training is admittedly out of date, but my understanding is that memory is allocated to the application as a whole, not multiple times to each separate thread. It is when each of those threads grabs 4Gb ram Yes, in posting the Windows screenshot, I made it clear that it was "not an exact comparison", but an example of the clear, factual, information that would enable us to understand your concerns and point you in the right direction.Ä¥ Rosetta tasks launch 15 parallel threads, a light workload. Where did you get your Masters in Computer Science? It is not a client problem, It is only on the Ryzen 7 box that I have to set max_ncpus_pct below 100 in. The AMD A6-9225 and AMD A6-5200 laptops in my render farm behave according to your perspective with no ludicrous system loads. If your system allows 4 threads, then 50 would only allow 2 thread. Pay attention: this only happens on my Ryzen box. 1 Look at /var/lib/boinc-client/globalprefs.xml /var/lib/boinc-client/globalprefsoverride.xml Override will override whatever is in globalprefs.xml Make adjustments such as changing maxncpuspct value from 50 to 100. Einstein, for example, launches 2 threads per task. Rosetta just grabs what resources Boinc allocates. Your method of counting threads, or /proc/cpuinfo: siblings, as cores, then launching that number of tasks is what generates 48 parallel threads competing for the 2 threads on each of the 8 cores. Active disc r/w competing with system swap on the same drive is what overloads things. It is when each of those threads grabs 4Gb ram that the system grinds to a crawl, taking minutes just to get a command prompt in a konsole window to kill the threads and cease using the swap partition. ![]() The Manager is a separate program and does not have to run all the time. It provides a graphical interface for monitoring and controlling the BOINC Client (which is sometimes also called the 'core client'). 5 Rosetta tasks launch 15 parallel threads, a light workload. The BOINC Manager is a control panel for BOINC. You don't seem to understand the difference between CPU cores and active threads. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |