| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Completed Research
Forum: FightAIDS@Home
Thread: Which batch is bad? |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 13
|
|
| Author |
|
|
uplinger
Former World Community Grid Tech Joined: May 23, 2005 Post Count: 3952 Status: Offline Project Badges: 
|
The batches that were built improperly were from 5008 to 5017. The batches with longer work units were unexpected but we are going to let them run through since they are only a few batches and it would take a lot more work on the work unit level to stop them. The batches that were built improperly will just error out quickly as they were missing a key file to run.
-Uplinger |
||
|
|
Rickjb
Veteran Cruncher Australia Joined: Sep 17, 2006 Post Count: 666 Status: Offline Project Badges: 
|
After the long faah50xx_ and faah499x_ series of WUs, I'm now getting normal-length faah418x_ and faah419x_ ones. BOINC initially underestimated the times for the long WUs and downloaded too many of them, and is now overestimating the times for the normal ones and downloading too few.
----------------------------------------I had thought that WCG put some kind of index into the WU headers to help BOINC calculate its estimates, but it seems that BOINC just uses some kind of averaging system. People who have an intermittent internet connection and run a short work queue may sometimes run out of WUs until the BOINC estimates stabilise, unless they increase the Workunit Cache setting in their WCG Device Manager. [Edit 1 times, last edit by Rickjb at Jul 29, 2008 4:01:28 AM] |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
1. After the long faah50xx_ and faah499x_ series of WUs, = BOINC initially underestimated the times for the long WUs and downloaded too many of them,
----------------------------------------2. I'm now getting normal-length faah418x_ and faah419x_ ones. = now overestimating the times for the normal ones and downloading too few. This is exactly how is works. The header of 1. used the running average normal flop count, thus shows too short time. By the time 2. arrived, the running average had gone up, thus the header carried that higher value and thus fetches less. Will take time again to lower the running average in the header. BOINC uses what's in the header to compute how long it's going to take and will adjust down the estimate with each completed 'normal' job.
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
|