Message boards :
Wish list :
Queued WU's
Message board moderation
| Author | Message |
|---|---|
|
Send message Joined: 5 Jul 09 Posts: 14 Credit: 28,617,272 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
It would make a great deal of difference to me to be able to ask for say 3 days (or better still a week's worth) of WU's to be downloaded at one time similar to the manner that work is distributed on WCG thus enabling me to run whilst not connected to the internet and down/upload during visits to site. It would in fact enable me to resume and use maybe 10 cards. I am sure others would benefit too such as those with flaky internet connections and am interested to know if this idea presents any major problems at your end. |
|
Send message Joined: 4 Apr 09 Posts: 450 Credit: 539,316,349 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
This project builds WUs from the results of previously completed WUs. To introduce a longer queue time would significantly delay completing each batch run. As a simple example if you increase the queue size by one day and there are 50 Wus in the chain for one batch you would be delaying the final result by 50 days. GDF - can you let us know some details this ... on average how many WUs make up a discreet batch of each of the sub-projects? I know in the past there was a max queue size (I think 5 days) instead of the current 2 WUs per card so they clearly made a decision to move to the current configuration. The bonus points for quick turnaround existed even back then to encourage short queues but apparently that was not enough. My guess would be that due to the high rate of errors and timeouts that it simply became too difficult to process enough batches in a reasonable timeframe with the volunteer computers and researchers available to this project. On average 50% - 75% of the WUs I process have previously been errored/aborted/timed out by someone else. If they increase the max queue they likely would also have to simulataneously increase the number of concurrent batches (so there would be enough WUs for everyone to crunch) but that dilutes the crunching power available for each individual batch which further increases the total runtime, which may make it necessary to create even more batches to keep enough WUs in the pipeline. Each cycle adds to the overall turnaround time and the cumulative effect is likely exponential. You crunching pal on team XtremeSystems, Snow Crash Thanks - Steve |
©2025 Universitat Pompeu Fabra