| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Active Research
Forum: Africa Rainfall Project
Thread: Work Available |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 3596
|
|
| Author |
|
|
alanb1951
Veteran Cruncher Joined: Jan 20, 2006 Post Count: 1332 Status: Recently Active Project Badges: 
|
Mike,
All classes of units are taking more than 13 days to validate on average, despite the relative deadlines. I presume this is because of the length of time that they take to download and possibly because the extremes are not necessarily going to the fastest machines as used to be the case with IBM. If the behaviour of various wingmen's systems I've seen associated with my ARP1 results is typical, more than 25% of the work units have at least one "No Reply" or "Not Started by Deadline"[1] response - an automatic delay of 6 days - and on some occasions the retry for that also goes No Reply or NSD, although that's not helped by the retries getting a shorter deadline (combined with the download issues...)Add to that the recent significant backlog in actually validating ARP1 work units and you have your 13-day delay... As you [and others] have frequently mentioned in the past, the optimum way to run ARP1 is to never allow it to download more tasks than the client should run at once - the reality is probably very different for most users :-( By the way, thanks for the report :-) Cheers - Al. [1] At WCG, Not Started by Deadline is just reported as an Error, recognizable by the combination of stderr.txt only containing the BOINC client version stuff and the return date being shortly after the due date. Some other BOINC projects manage to mark that state explicitly in their result pages :-) |
||
|
|
Mike.Gibson
Ace Cruncher England Joined: Aug 23, 2007 Post Count: 12594 Status: Offline Project Badges:
|
Sunday Report .
30,857 units validated in 7 days, so average of 4,408.1 per day. This is up more than a third on last week. Assuming that a full generation 182 will be the last, there are 1,819,863 units still outstanding, so my forecast end date would now be 18 November 2023, however, we are still in testing so we should finish well before then. 2 of the 3 ultra extremes have moved in the last week, but need to be boosted. The definitions of Extreme or Accelerated units are unchanged. . There are now 108 Extremes and 172 Accelerated units, although the numbers in their generations are 112 & 2,142 due to lack of movement. The situation with downloading units is much improved, but still takes time. Validation time has improved significantly. Mike |
||
|
|
Mike.Gibson
Ace Cruncher England Joined: Aug 23, 2007 Post Count: 12594 Status: Offline Project Badges:
|
Al
In a perfect world 1 unit per usable thread would be ideal, but crunchers don't want to risk going without. I would suggest 1 per usable thread plus 1 spare would be a good compromise. For someone with 8 threads using 4 for ARP, the spare would allow up to 6 hours for a replacement to download. That should be plenty in normal operation. With the limited supply we have had during testing it has been human nature to carry a larger stock. The reduced validation times indicate a much better situation. Mike |
||
|
|
Unixchick
Veteran Cruncher Joined: Apr 16, 2020 Post Count: 1312 Status: Offline Project Badges: 
|
Thanks for the update Mike. It's good to see some good signs of progress. not perfect yet, but progress.
Can't wait to see how much I've crunched during the testing phase. I wonder if I can reach the next badge level before it finishes. |
||
|
|
mikey
Veteran Cruncher Joined: May 10, 2009 Post Count: 826 Status: Offline Project Badges: 
|
With such little bandwidth available, computers running scripts such as that make the Krembil servers see a DDOS attack. :-D That's funny. Our hosting provider did contact us about some suspected DDOS attacks. Can't say it was you guys for sure, though.  It could also be people asking for tasks with a cron job in Linux or using task scheduler in Windows, with a 2 min 30 sec backoff everytime it would be pretty easy to set one up to check for new tasks at the interval   |
||
|
|
mikey
Veteran Cruncher Joined: May 10, 2009 Post Count: 826 Status: Offline Project Badges:
|
Al In a perfect world 1 unit per usable thread would be ideal, but crunchers don't want to risk going without. I would suggest 1 per usable thread plus 1 spare would be a good compromise. For someone with 8 threads using 4 for ARP, the spare would allow up to 6 hours for a replacement to download. That should be plenty in normal operation. Mike For some of us that 6 hour download time is not nearly enough, some of us are experiencing retries every 5+ hours and then it happening again when that doesn't work What does work is banging on the door ie clicking retry though the latest thing is for WCG to just flat stop sending a task at some point and then enforcing a 4 min 30 sec timeout before even switching over to letting you click to retry the download |
||
|
|
Unixchick
Veteran Cruncher Joined: Apr 16, 2020 Post Count: 1312 Status: Offline Project Badges:
|
The hopper is empty. Can someone load in some more ARP WUs??
I'm guessing we will have to wait until Monday. Hopefully I'll get some resends to keep me busy until then. I'm trying to remember what the criteria is for getting the extreme WUs,...I think it was that you had to have a record of returning WUs witthin 48 hours ?? I think I've messed up my record recently, but hopefully I can earn it back quickly. |
||
|
|
Mike.Gibson
Ace Cruncher England Joined: Aug 23, 2007 Post Count: 12594 Status: Offline Project Badges:
|
Sunday Report .
18,664 units validated in 7 days, so average of 2,666.3 per day. This is down 40% on last week. Then there were catch ups. Assuming that a full generation 182 will be the last, there are 1,801,199 units still outstanding, so my forecast end date would now be 14 August 2024, however, we are still coming out of testing so we should finish well before then. None of the 3 ultra extremes have moved in the last week. Have they stuck again? Do they need their timestep adjusted? The definition of Accelerated units has moved on 1 generation. . There are now 102 Extremes and 164 Accelerated units, although the numbers in their generations are 106 & 2,304 due to lack of movement. The situation with downloading units still takes time. Validation time is improving. Mike |
||
|
|
mikey
Veteran Cruncher Joined: May 10, 2009 Post Count: 826 Status: Offline Project Badges:
|
Sunday Report . 18,664 units validated in 7 days, so average of 2,666.3 per day. This is down 40% on last week. Then there were catch ups. Assuming that a full generation 182 will be the last, there are 1,801,199 units still outstanding, so my forecast end date would now be 14 August 2024, however, we are still coming out of testing so we should finish well before then. None of the 3 ultra extremes have moved in the last week. Have they stuck again? Do they need their timestep adjusted? The definition of Accelerated units has moved on 1 generation. . There are now 102 Extremes and 164 Accelerated units, although the numbers in their generations are 106 & 2,304 due to lack of movement. Validation time is improving. That's all very good News, Thanks Mike!! The situation with downloading units still takes time. Mike I sure hope they are working on this instead of just saying 'it's okay as it is'!! mikey [Edit 1 times, last edit by mikey159b at Oct 11, 2022 12:47:27 AM] |
||
|
|
catchercradle
Senior Cruncher England Joined: Jan 16, 2009 Post Count: 171 Status: Offline Project Badges: 
|
Hopefully I'll get some resends to keep me busy until then. Eventually got a resend, a _5 which may be a personal record going for the 6th attempt at a work unit. |
||
|
|
|