| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Active Research
Forum: Africa Rainfall Project
Thread: Not enough time given for a task |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 102
|
|
| Author |
|
|
uplinger
Former World Community Grid Tech Joined: May 23, 2005 Post Count: 3952 Status: Offline Project Badges: 
|
Hello Billy,
As Mike.Gibson has mentioned, these work units are pretty intensive. It is one of our harder projects in terms of computational time and check point splits compared to our other projects. Each work unit sent can range in time even on a single machine, this is due to moisture levels and other variables in the area/grid that the individual work unit is searching for. Also, as Mike has mentioned, even if your work unit is late, we try to grant credit. Credit is still granted within 24 hours of the entire work unit being validated. What this means is that if 2 copies were sent out and one of them did not finish in time for the deadline. An additional result would be sent out. If that third result and the first result validate against each other. The time allowed for the second result to still get credit would be 24 hours after the 1st and 3rd result validated. Mike, Thanks for your responses. As for your question on the 30's, I have a prioritization method for older generations. It goes as follows based off max generation per batch (each batch is 100 work units, ARP1_0025006_029 -> batch ARP1_00250). In this case, the generation is 29. If max generation for that batch is 29, then priority is 0 (normal). If generation is 1 less than max generation, priority is 0. If generation is 2 less than max generation, priority is 1 (just means it gets pushed to the top of the load queue, runs normal for users), if generation is 3 or more behind max generation, then priority is set to 10 (This invokes the reliable hosts only). It is a way to help catch the work units up and allows the researchers to do evaluations as they need all from a given generation to do analysis. Hope this helps and thanks to everyone for supporting us, -Uplinger |
||
|
|
Aurum
Master Cruncher The Great Basin Joined: Dec 24, 2017 Post Count: 2384 Status: Offline Project Badges: 
|
The real problem here is people like Rod above (RTS48) who have too large a cache to be able to complete the work in time, At least Rod seems to have got the message but there are many more of them out there. Hopefully some of them may have seen those messages. Mike I also use my 4 Venues to fine tune some more. If I had 8 Venues I'd use them all to do it even more efficiently. In fact, I wish I could associate my work queue with each individual Device. E.g., .............ARP...OPN Default.....0.....44...(E5-2683 v3) Home.......9.....11...(i7-6950X) Work......17.....30...(i9-9960X) School....21.....23...(E5-2699 v4) I also use app_config.xml to limit running ARP WUs to [(t/2)-1] so they run almost as fast as disabling hyperthreading to avoid the ARP L3 Cache congestion. I'd like to say something like it's best to run ARPs on CPUs with BOINC CPU FP Benchmarks over 3,300 and Integer Benchmarks over 80,000 but something changed and the benchmarks are a mess now. I don't know if it was upgrading from Linux Mint 19.3 to LM 20 or upgrading Linux BOINC client from 7.9.3 to 7.16.6 which happened at the same time. Integer BMs were cut almost in half. Also Windows Integer BMs are around a quarter of Linux. ARP WUs run great and I haven't seen a single Invalid one yet. I still hate the long checkpoints but I adapted and plan for a reboot 3 to 4 hours ahead.   ...KRI please cancel all shadow-banning[Edit 2 times, last edit by Aurum420 at Oct 22, 2020 12:45:58 AM] |
||
|
|
Falconet
Master Cruncher Portugal Joined: Mar 9, 2009 Post Count: 3295 Status: Offline Project Badges: 
|
It's was the change from Ubuntu 18.04 to Ubuntu 20.04 that caused the difference in benchmarks. Haven't seen any decrease in performance from the change.
----------------------------------------AMD Ryzen 5 1600AF 6C/12T 3.2 GHz - 85W AMD Ryzen 5 2500U 4C/8T 2.0 GHz - 28W AMD Ryzen 7 7730U 8C/16T 3.0 GHz |
||
|
|
Billy Ewell 1931
Cruncher Joined: Mar 1, 2008 Post Count: 22 Status: Offline Project Badges:
|
My special thanks to Mike Gibson and WCG Tech Uplinger for some most-informative observations and advice.
As best as I can determine, I had downloaded several tasks for 'buffering' as a participant in the "THOR" challenge underway on WCG and, when I included the Africa Rainfall Project in those tasks to be processed, I subsequently (probably) did not remove ALL the non-ARP tasks from the cache and apparently that resulted in the inability of my 2.66 GhZ machine from finishing the ARP task within the established time limit. But obviously the time to complete this one ARP task would have taken considerable more hours than was indicated on Boinc Manager. I did however follow the same procedures in installing ARP tasks on my other Five (5) computers (i3s,i7,Xeons and Notebook i3) and all tasks completed successfully. Anyway, I again emphasize the appreciation of the inputs from Mike and Uplinger and I withdraw my over-reacted criticism. What I want to add in a different vain is the appreciation and admiration I have for World Community Grid and the fact that IBM Corporation did a tremendously fine undertaking for the world and its occupants, both human and animals, etc., when it chose to underwrite this Scientific Research entity. Bill (Celebrating 89 years on Earth in November)  |
||
|
|
Mike.Gibson
Ace Cruncher England Joined: Aug 23, 2007 Post Count: 12160 Status: Offline Project Badges: 
|
Bill
2 things to prevent on arp1. The first is not to utilise more than half of each machines threads on arp1. That can be done using app_config.xml. The other is to keep the cache for arp1 below half the deadline times for each machine. That can be done in Project Limits in Device Profiles and is common to all projects. Also mip1 should be restricted to one third of each machines threads, again using app_config.xml. Let me know if you need any help with those. You are only 11 years ahead of me. Mike |
||
|
|
RTS48
Veteran Cruncher Bolivia Joined: Aug 2, 2009 Post Count: 1350 Status: Offline Project Badges: 
|
After all of my problems of work not completing I reduced my Cache (see above) and lo and behold - all of my problems have evaporated. It is really heartening that there is so much expertise and so many helpful people here in the forums. Once again, thank you all.
----------------------------------------
Rod Peel
Santa Cruz Bolivia South America  ,  |
||
|
|
KerSamson
Master Cruncher Switzerland Joined: Jan 29, 2007 Post Count: 1671 Status: Offline Project Badges: 
|
Hi Rod,
----------------------------------------you're very welcome. Yves |
||
|
|
Sgt.Joe
Ace Cruncher USA Joined: Jul 4, 2006 Post Count: 7594 Status: Offline Project Badges:
|
My kids get tired of me telling them I started college using a slide rule and graduated before the PC had been invented. I think that's why my generation can do math in our heads. I still have slide rule, although I haven't used it in a while. I have brought it to school when I am volunteering or substitute teaching (all out the window now with Covid-19 for a while). I may not be able to all the math in my head, but at least I can estimate well so I know my answer on the calculator is within the correct bounds. OK, this was off topic, so back to the amount of time for ARP. Cheers
Sgt. Joe
*Minnesota Crunchers* |
||
|
|
Mike.Gibson
Ace Cruncher England Joined: Aug 23, 2007 Post Count: 12160 Status: Offline Project Badges:
|
I still have the slide rule that I had at school. I also programmed my office desktop machine (Olivetti Programma) in 1967. It could hold all of 128 instructions at one time (but we could use multiple magnetic cards to enlarge the programs).
Mike |
||
|
|
|