| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Active Research
Forum: Africa Rainfall Project
Thread: All tasks failing on linux host |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 38
|
|
| Author |
|
|
alanb1951
Veteran Cruncher Joined: Jan 20, 2006 Post Count: 873 Status: Offline Project Badges: 
|
I have an i9 with 20 threads and 40 GB of RAM, it's a fix computer Given the specs on your computer, their really should be no way the ARP work units should take 36 hours. If you are running all twenty threads with ARP units, I would suspect your computer is doing some self throttling due to heat issues. If that is not the case, you must have some other bottleneck in the processing stream someplace. I have an I7-7700 with 8gb RAM and ARP units take 18-20 hours. I only run 2 threads on ARP, the rest are MCM. I am curious to know what a "fix" computer is.  Cheers It isn't such an issue for server-type CPUs with their extra memory-handling (and, typically, lower CPU clock frequencies anyway), but it will have a throttling effect on other processors simply because the instruction pipelines will stall more often! -- think back to the MIP1 project (which used a Rosetta app, I believe) where the more MIP1 tasks one ran the slower they ran and the cooler the CPU cores ran! Another possible bottleneck is disk access; if that system uses SSDs it won't show up much, but if using spinning rust the amount of I/O traffic from an ARP1 task is quite high, even between checkpoints... That said, 36+ hours still seems high -- however, without actual access to the user's system to see configuration, other workload, et cetera, it's not something we are going to be able to fathom out :-) Cheers - Al, P.S. -- my old i7-7700 (16GB) typically takes 11 or 12 hours to run ARP1 tasks (never more than two, and never more than 6 "CPUs" in use for BOINC...); my newer Ryzens all do quite a bit better (more L3 cache, better memory access, more instruction pipelines, etc.) |
||
|
|
Sgt.Joe
Ace Cruncher USA Joined: Jul 4, 2006 Post Count: 7579 Status: Recently Active Project Badges: 
|
Out of the 8 threads on the I7700, I only run 2 threads of ARP and the rest MCM, so I do keep it fully loaded. However, I see no degradation of any thing I am trying to do. So BOINC is happily staying in the background and not bothering other work.I am definitely not memory bound as I am using a slight bit more than 4gb out of the 8.
----------------------------------------Cheers
Sgt. Joe
*Minnesota Crunchers* |
||
|
|
[AF>Le_Pommier] Jerome_C2005
Cruncher Joined: Aug 17, 2006 Post Count: 29 Status: Offline Project Badges: 
|
That said, 36+ hours still seems high -- however, without actual access to the user's system to see configuration, other workload, et cetera, it's not something we are going to be able to fathom out :-) Intel i9-10910 CPU @ 3.60GHz on an late 2020 iMac. So yes, definitively not a "high end crunching config", but a loyal and tough cookie crunching 24H/7D full throttle. One could wonder why take such a machine to crunch ? But I say : never underestimate the WAF . |
||
|
|
alanb1951
Veteran Cruncher Joined: Jan 20, 2006 Post Count: 873 Status: Offline Project Badges:
|
That said, 36+ hours still seems high -- however, without actual access to the user's system to see configuration, other workload, et cetera, it's not something we are going to be able to fathom out :-) Intel i9-10910 CPU @ 3.60GHz on an late 2020 iMac. So yes, definitively not a "high end crunching config", but a loyal and tough cookie crunching 24H/7D full throttle. One could wonder why take such a machine to crunch ? But I say : never underestimate the WAF . Temporarily restrict the system to only use 10 "CPUs" and see how fast the tasks run. I suspect you'll see a major improvement, and if tasks run more than twice as fast that's a definite win! :-) If you do try that, please report back on the success (or otherwise)... Cheers - Al. P.S. I presume you are currently running with both CPU-related usage figures set to 100%? Changing the one that controls CPU time rather than cores used is not a good idea :-) [Edit 1 times, last edit by alanb1951 at Nov 13, 2024 7:11:59 PM] |
||
|
|
Mike.Gibson
Ace Cruncher England Joined: Aug 23, 2007 Post Count: 12142 Status: Offline Project Badges: 
|
The usual reccommendation here is to run a maximum of 50% of threads (using app_config and the rest on whatever else is available which currently is MCM only.
There is a very large amount of file movement involved in ARP, especially at the checkpoints which occur every 12.5%. Try spreading the units out. by short suspensions. so that the checkpoints do not occur at the same time. Mike |
||
|
|
catchercradle
Advanced Cruncher Joined: Jan 16, 2009 Post Count: 99 Status: Offline Project Badges: 
|
The usual reccommendation here is to run a maximum of 50% of threads (using app_config and the rest on whatever else is available which currently is MCM only. With ARP I find maximum throughput is using N-1 threads where N is the number of real cores rather than threads. Using virtual cores share a floating point unit which is why using more threads than real cores slows things down. The same happens with CPDN which is not surprising as there are shared elements in the task software. |
||
|
|
[AF>Le_Pommier] Jerome_C2005
Cruncher Joined: Aug 17, 2006 Post Count: 29 Status: Offline Project Badges:
|
I get the idea to run without HT but
----------------------------------------- I can't globally deactivate this under macOS, unlike (most) PCs - reducing boinc to 50% of possible thread is no guarantee that it will really work like this (i.e. on physical cores) since there are always other running processes at the same time (system... even with small amount of CPU used), so in my views it will just "increase the possibility" that the physical core are used instead of HT thread But it might be worth trying, I'll give it a try, but not soon due to team targets at the moment, hopefully next year. |
||
|
|
Mike.Gibson
Ace Cruncher England Joined: Aug 23, 2007 Post Count: 12142 Status: Offline Project Badges:
|
I seem to get by by limiting my 4 core 8 threads to a maximum of 4 ARP1 with app_config. The remainder are filled by MCM1 which are tiddlers by comparison.
Mike |
||
|
|
|