Message boards :
Number crunching :
Bigger work buffer
Message board moderation
Previous · 1 · 2
Author | Message |
---|---|
W-K 666 Send message Joined: 18 May 99 Posts: 19072 Credit: 40,757,560 RAC: 67 |
% of time BOINC client is running 71.8875 % The thing is, if you have any VHAR units in there that have deadline if 8 days then the BOINC manager is going to think because of 10 day connection interval that they will have difficulty in being returned on time. Therefore no more work, until they are returned. To allow for any unexpected delays in uploading and reporting the manager will try to complete units 24 hrs before the actual deadline. Andy |
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
% of time BOINC client is running 71.8875 % Several people have suggested that you might try a shorter connect interval (i.e. less than one day). The connect interval interacts with cache size in ways you might not expect. ... but if you won't try them, even as an experiment, then we can't help. |
Rowe Family and Friends Send message Joined: 25 Dec 00 Posts: 17 Credit: 38,395,231 RAC: 67 |
i have and it makes no difference i cant get anymore than about 20 work units at once. the thing is it works fine changing the cashe size on my desktop. i change it from 5/5 to 10/10 and it downloads heaps of units, yet my laptop doesnt get any ill leave it and see what it does over the weekend |
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
i have and it makes no difference i cant get anymore than about 20 work units at once. the thing is it works fine changing the cashe size on my desktop. i change it from 5/5 to 10/10 and it downloads heaps of units, yet my laptop doesnt get any You're using numbers like 5 and 10, when I'm suggesting values below 1. |
Rowe Family and Friends Send message Joined: 25 Dec 00 Posts: 17 Credit: 38,395,231 RAC: 67 |
i have and it makes no difference i cant get anymore than about 20 work units at once. the thing is it works fine changing the cashe size on my desktop. i change it from 5/5 to 10/10 and it downloads heaps of units, yet my laptop doesnt get any thats cuase i was talking about my desktop like i said it hasnt made any difference having the cashe setting at .1/10 or .5/5 or anything like that i still only get about 20 units |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14653 Credit: 200,643,578 RAC: 874 |
i have and it makes no difference i cant get anymore than about 20 work units at once. the thing is it works fine changing the cashe size on my desktop. i change it from 5/5 to 10/10 and it downloads heaps of units, yet my laptop doesnt get any If making all those changes is having no effect at all, then the change isn't being recognised by the laptop. Why not? 1. Not updating client after making preference change (unlikely, but included for completeness) 2. Not updating correct 'venue' preference for computer 3. Preference override file on computer negates all changes Got to be one of those - check all settings (including those you know couldn't possibly be wrong, LOL) and tell us what you see. |
Alinator Send message Joined: 19 Apr 05 Posts: 4178 Credit: 4,647,982 RAC: 0 |
If making all those changes is having no effect at all, then the change isn't being recognised by the laptop. Why not? Hmmm... Take another look at the host's time metrics. From that I would expect it to only carry about half of what you would think it should from the cache settings. Also, it hasn't been mentioned here yet, but when you run decoupled the CI doesn't play a role in determining the cache size anymore per se. IOW. setting BOINC to 10/10 doesn't mean you will carry 20 days worth of work. As John has tried to explain, the difference between coupled cache mode and decoupled is that in coupled mode the CC will make sure that all deadlines can be met by at least one contact session sooner than when they are ultimately due back and adjust work fetching and scheduling accordingly. When running decoupled, the CC ignores the impact of any contact schedule when deciding how much work to get and when to run it and only considers the deadlines of the tasks onboard. This doesn't make for much of a difference if the host has unrestricted network access and/or runs 24/7, but can lead to missed deadlines for dialup hosts, part timers, and/or ones where you have set a restricted network schedule when running decoupled. Alinator |
John McLeod VII Send message Joined: 15 Jul 99 Posts: 24806 Credit: 790,712 RAC: 0 |
If making all those changes is having no effect at all, then the change isn't being recognised by the laptop. Why not? 10 days Connect Interval and 10 days of extra work would indeed be an attempt to keep 20 days worh of work on the host. The Connect interval is used to both determine the minimum cache, and the calculation of the task completion deadlines (work must be completed before one connect interval before the report deadline). If the computer has uninterupted network access, you are best of with a CI of 0. Use the extra work setting to maintain a queue of work. BOINC WIKI |
Alinator Send message Joined: 19 Apr 05 Posts: 4178 Credit: 4,647,982 RAC: 0 |
10 days Connect Interval and 10 days of extra work would indeed be an attempt to keep 20 days worh of work on the host. The Connect interval is used to both determine the minimum cache, and the calculation of the task completion deadlines (work must be completed before one connect interval before the report deadline). If the computer has uninterupted network access, you are best of with a CI of 0. Use the extra work setting to maintain a queue of work. Well I stand corrected on that. I must have misunderstood you when we we're talking about that when the feature came about. When I tried testing the 10/10 20 day hypothesis on one of my fastest hosts, and I never got over what I would carry in the cache in coupled mode, so I assumed you had addressed the matter of irresolveable scheduling paradoxes when running decoupled by having the CI just set the work fetch and report trigger point and nothing else. Based on this info, you can still shoot yourself in the foot pretty good if you go 10/10 and don't pay attention to what's going on under some circumstances. ;-) Alinator |
John McLeod VII Send message Joined: 15 Jul 99 Posts: 24806 Credit: 790,712 RAC: 0 |
10 days Connect Interval and 10 days of extra work would indeed be an attempt to keep 20 days worh of work on the host. The Connect interval is used to both determine the minimum cache, and the calculation of the task completion deadlines (work must be completed before one connect interval before the report deadline). If the computer has uninterupted network access, you are best of with a CI of 0. Use the extra work setting to maintain a queue of work. 12 guage anyone? BOINC WIKI |
recondas Send message Joined: 3 Nov 06 Posts: 2 Credit: 58,756 RAC: 0 |
I've seen a number of convincing arguements for both large and small buffer sizes. Unfortunately, it looks like I've got a work unit pending where my wingman is the poster child for too much buffer. Looking at the listed computer, I'm not sure how he managed to accumulate *2157* tasks, but dozen's of them have died on the vine, and more, including mine probably aren't far behind. |
kittyman Send message Joined: 9 Jul 00 Posts: 51468 Credit: 1,018,363,574 RAC: 1,004 |
I've seen a number of convincing arguements for both large and small buffer sizes. How in the world is a doggy Celery processor sitting on that much work?? "Freedom is just Chaos, with better lighting." Alan Dean Foster |
recondas Send message Joined: 3 Nov 06 Posts: 2 Credit: 58,756 RAC: 0 |
How in the world is a doggy Celery processor sitting on that much work?? I scrolled thru all 20+ pages of work units. It looks like the last one processed was in mid September - but the system continued to download work on a nearly daily basis since (as recently as yesterday). There may be good reasons why processing stopped, but it is rather surprising the system continued to request work. Makes one wonder what the quota settings are. |
web03 Send message Joined: 13 Feb 01 Posts: 355 Credit: 719,156 RAC: 0 |
How in the world is a doggy Celery processor sitting on that much work?? At least it's down to a 1 per day quota. EDIT: But is taking 37 days to turn anything around. |
Keith T. Send message Joined: 23 Aug 99 Posts: 962 Credit: 537,293 RAC: 9 |
According to BOINC Stats that host does occasionally return a valid WU. http://boincstats.com/stats/host_graph.php?pr=sah&id=2415609 Why is it hoarding so many? I can't work out if this is being done accidentally or deliberately ! Sir Arthur C Clarke 1917-2008 |
DJStarfox Send message Joined: 23 May 01 Posts: 1066 Credit: 1,226,053 RAC: 2 |
According to BOINC Stats that host does occasionally return a valid WU. That sounds like a BOINC server feature request! "Absolute max number of sent WU per computer". This would limit downloaded tasks to a specific number, until either the machine returns valid result or a valid result past its deadline. |
kittyman Send message Joined: 9 Jul 00 Posts: 51468 Credit: 1,018,363,574 RAC: 1,004 |
It doesn't really matter much in the total context of things. If one host has a few thousand extra WUs, the rest of the system goes on....... Hangs up a few results, but it's no biggy. The whole thing about pending WUs is getting stale. They have been increasing, and they will increase until they reach an apex and then they will settle down. If you put 38,000 in pending like I have, and then it stays about the same, it's like you are getting your results granted in real time, they shouldn't build any further. And if they do, for a short span, they will come back down and things will even out. It's all just a puff of smoke, and should not concern anybody. The real concern is whether the servers can be configured to stay stable for more than a few days at a time, whether Seti can secure enough funding to stay running, and whether Arecibo comes back and stays on line to feed Seti the data that we are all analyzing. Any of the above fails, and we are all without our pet project. "Freedom is just Chaos, with better lighting." Alan Dean Foster |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.