| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: BOINC Agent Support
Thread: Delayed polling for more WU's?? |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 12
|
|
| Author |
|
|
gb077492
Advanced Cruncher Joined: Dec 24, 2004 Post Count: 96 Status: Offline |
Interesting. I've got "Connect about every" set to 0.00 days and "Additional work buffer" set to 0.40 days (on the assumption that outages are unlikely to last longer than one 8-hour shift) on all my machines, including an 8-way. As I said, with these settings I've had no issues at all.
Is there any reason why you should not have "Connect about every" set to 0.00 days too? |
||
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges: 
|
Interesting. I've got "Connect about every" set to 0.00 days and "Additional work buffer" set to 0.40 days (on the assumption that outages are unlikely to last longer than one 8-hour shift) on all my machines, including an 8-way. As I said, with these settings I've had no issues at all. Is there any reason why you should not have "Connect about every" set to 0.00 days too? If you've got a permanent internet-connected, using zero for "Connected..." is the recommended thing to do **. If you on the other hand is not permanently connected, it's recommended to set "Connected..." to be a little larger than your expected offline-time, due to the rules below. In the current v6.10.xx-clients, the usage of "Connected..." is these: 1st. rule is, your cache-size is the sum of "Connected..." and "Additional...", so if example "Connected..." is 1 day and "Additional..." is 3 days, you should normally have 4 days cached. 2nd. rule is, if "Connected..." > "deadline", you won't get assigned this task, except idle cpu. So, with WCG-work having a 10-day deadline, if you set "Connected..." to 11 days or something, you won't get any work from WCG except idle cpu. 3rd. rule is, if at all possible, the client-scheduler makes sure all work is finished atleast "Connected..." before the deadline, and any finished task will be reported if less than "Connected..." until deadline (if BOINC has got an usable network-connection). The 4th. rule only applies if you're running multiple BOINC-projects. If a project has LTD < - "Additional...", this project won't be asked for work, except idle cpu. The client will on the other hand ask all contactable projects (except zero-resource-share projects), sorted by LTD *, if cached work on client is less than "Connected...". Older clients has different rules, example rule-4 was new in the v6.6.xx-clients, and v5.10.13 and earlier did report if more than "Connected..." since task finished... So, does current clients follow these rules? They should do, but it's possible there's still bugs in v6.10.58, so in some instances they're violated... The most common reasons for no work-requests are connection-problems, client being backed-off work-requests, and estimated run-times being off. * LTD = Long term debt, also called "resource-type work fetch priority", can be seen by selecting a project and hitting "preferences". ** The full preference-text is: "Computer is connected to the Internet about every ... days Leave blank or 0 if always connected. BOINC will try to maintain at least this much work."  "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." |
||
|
|
|