| 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 |
|
|
Mike.Gibson
Ace Cruncher England Joined: Aug 23, 2007 Post Count: 12594 Status: Offline Project Badges: 
|
Earlier today I received my first double re-send (suffix 3). The zero errored after only 2 minutes and the fist re-send (2) detached after 35 hours. I was still allocated the full 7 days for returning it.
Mike |
||
|
|
KerSamson
Master Cruncher Switzerland Joined: Jan 29, 2007 Post Count: 1684 Status: Offline Project Badges: 
|
Since the last 2 days, I've just fetched very few ARP WUs.
----------------------------------------Fetching ARP WUs remains very challenging. With a daily average of 60 CPU days with relatively good performing machines, I cannot understand why I earns only about 270 days of ARP work thought other members with half CPU days already reach 2 or more years. I have the feeling that something in the middle is unbalancing the distribution between the members. Definitively, after 4 months, the differences are more than significant. Happy crunching, Yves |
||
|
|
uplinger
Former World Community Grid Tech Joined: May 23, 2005 Post Count: 3952 Status: Offline Project Badges: 
|
Greetings everyone,
We have had a discussion with the researchers and are increasing the speed of the project to send out 4,000 work units per day ( 8,000 results ). I have made the change on our server this morning and additional work units will be flowing within the next 30 minutes. Thanks, -Uplinger |
||
|
|
KerSamson
Master Cruncher Switzerland Joined: Jan 29, 2007 Post Count: 1684 Status: Offline Project Badges:
|
Thanks Uplinger.
----------------------------------------But nothing fetched excepted more MCM and MIP  |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Greetings everyone, We have had a discussion with the researchers and are increasing the speed of the project to send out 4,000 work units per day ( 8,000 results ). I have made the change on our server this morning and additional work units will be flowing within the next 30 minutes. Thanks, -Uplinger Better start watching how the hamstering is going to affect the return times. With greater availability maybe some come to the insight of reducing the profile setting. Remember guys, a unit simulates 48 hours weather evolution only and there being about 183 iterations with 24 hours computing time every day a return takes longer increases project duration by a lot. |
||
|
|
Mike.Gibson
Ace Cruncher England Joined: Aug 23, 2007 Post Count: 12594 Status: Offline Project Badges:
|
Yves
I don't know what your settings are but you seem to have a lot more threads/cores than I have (8 threads on one machine only) and we both have Ruby. Provided that you stick to half arp and half mcm/mip, with mip restricted to 3/8 on each machine, you should get plenty of arp. Even more now that the flow has doubled again. See earlier posts of mine for my full strategy. Mike |
||
|
|
KerSamson
Master Cruncher Switzerland Joined: Jan 29, 2007 Post Count: 1684 Status: Offline Project Badges:
|
Mike,
----------------------------------------I followed the recommendations provided over the last two months. Limited number of MCM and MIP WUs, maintaining the setup buffer relatively empty. Just last night, I finally fetched about 20 ARP WUs, it is the first time I will have more than 1 or 2 per machine. I do not "hamster" ARP WUs, all of the fetched ARP WUs are currently being computed. Cheers, Yves |
||
|
|
Mike.Gibson
Ace Cruncher England Joined: Aug 23, 2007 Post Count: 12594 Status: Offline Project Badges:
|
The principle is that the more arp units you crunch at any one time on any one machine, the less often that machine requests more work, because of the length of crunching time. It is the reporting of mcm in particular that enables you to catch more arp when they are available.
Also the more arp or mip you crunch, the slower your machine goes. Hence the suggestion to limit arp to half the threads available and to limit mip to less than mcm. Then there is the aspect of keeping your machines listed as 'reliable' which means fast turn around. The arp buffer should be limited to one extra unit per thread available for arp. That means when you are restricting to half the threads then the total held should not exceed the total number of threads. Mike |
||
|
|
KerSamson
Master Cruncher Switzerland Joined: Jan 29, 2007 Post Count: 1684 Status: Offline Project Badges:
|
Hi Mike,
----------------------------------------I am completely aware about all the points you mentioned. Even with mostly MCM, FAH2 and MIP, my machines - excepted last night - never fetch more than 1 or 2 ARP WUs each, if any. However, I know that I need to be patient, even if I don't understand exactly why it is as it is: wait and see. Yves |
||
|
|
Jack007
Master Cruncher CANADA Joined: Feb 25, 2005 Post Count: 1604 Status: Offline Project Badges:
|
I don't run at max capacity,
----------------------------------------5 MIP and I think unlimited for ARP (and HSTB but they are rare) I have been getting 4 to 8 ARP fairly consistently now. So much so I gotta repair another computer :D  |
||
|
|
|