Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
World Community Grid Forums
Category: Completed Research Forum: Discovering Dengue Drugs - Together - Phase 2 Forum Thread: Keep getting high priority jobs: can't fill up cache... [RESOLVED] |
No member browsing this thread |
Thread Status: Active Total posts in this thread: 13
|
Author |
|
bicotz
Advanced Cruncher Canada Joined: Apr 25, 2010 Post Count: 67 Status: Offline Project Badges: |
Hello all,
----------------------------------------My most trusted machines keep getting high priority jobs. The downside to this is that it's preventing the machines from requesting new jobs, thus my caches are very low... If you suspend a "normal" job, you will also get this type of behavior. Is there a way around this; to either quit receiving Hi priority jobs or find a way around to fill my caches? The only way I can think of is to error out a WU either by overclocking to high or cutting off power sudenly, that way the machine won't be considered reliable... That's what happened on two of my machines and since then, they keep getting regular WU's only. Thanks for any response! [Edit 1 times, last edit by bicotz at Dec 24, 2010 2:29:27 AM] |
||
|
KWSN - A Shrubbery
Master Cruncher Joined: Jan 8, 2006 Post Count: 1585 Status: Offline |
As far as I know, yes, that would be the only way. You can always abort the high-priority tasks if you are that desperate to fill your cache, these jobs will get passed on to the next cruncher with a reliable machine.
----------------------------------------As for ethics, I'll let you decide. Distributed computing volunteer since September 27, 2000 |
||
|
gb009761
Master Cruncher Scotland Joined: Apr 6, 2005 Post Count: 2977 Status: Offline Project Badges: |
Hi bicotz, I'd question as to why you want your cache filled up - as, providing that your computer has a WU to crunch on all available cores, then what's the issue. As soon as a WU is finished, BOINC will realise that it needs to request some more work and whether that's a High Priority WU or a normal one (when it'll fetch more), then what's the problem.
---------------------------------------- |
||
|
anhhai
Veteran Cruncher Joined: Mar 22, 2005 Post Count: 839 Status: Offline Project Badges: |
when you say high priority WU, do you mean WUs that have the words 'high priority" next to them or repair WUs? If you meant repair WU, then they should have no effect on you filling up your queues. However, you are most likely referring to WUs with the words "high priority" next to them when they are running. There is no difference between a "high priority" and a normal WU, except for the fact that BOINC thinks that you may not finish the WU by the deadline.
----------------------------------------Based on your problem, my guess is that you have not been running BOINC for a while or BOINC doesn't run 24/7 on your machine. BOINC keeps track of what percentage of time your computer is on and what percentage BOINC is running while your computer is on. This combine with the speed of your system and past performace allows it to figure out how long it would take you to finish a WU. For example, the WU normally takes 4hr of CPU time to run on your system, but your computer is on only 50% of the time (8am-8pm) and BOINC is running 25% of the time your computer is on (run time doesn't include suspended periods). So BOINC will figure that the 4hr WU will take 32 hrs to finish (4/.5/.25 = 32) This also happens when you go on vacation and turn off BOINC for a few days. Leave BOINC running and it will realize that your system can finish WU faster and it will start asking for more WU to fill you cache. |
||
|
mwgiii
Advanced Cruncher United States Joined: Aug 17, 2006 Post Count: 131 Status: Offline Project Badges: |
I agree with anhhai.
----------------------------------------One of my machines is doing the same thing. Just let the WUs finish. Your estimated time to completion for your WUs will adjust and everything will return to normal soon. |
||
|
bicotz
Advanced Cruncher Canada Joined: Apr 25, 2010 Post Count: 67 Status: Offline Project Badges: |
Problem solved! Here's the explanation: a week ago I was doing some testing and upped the cache on some of my computers up to @ 7 days. These machines run 24/7. So far no prob, DDDT2 leaves 10 days to complete the WU. The problem arose once I started receiving LOTS of "PE" units (resends it seems) whit a short dealine (2-3 days) that took 12 hours to complete. These units somehow backlogged the other units in the cache so much that the computer realised that the 10 day deadline could not be met... So all the older units, as soon as work is started, become "running high priority"
----------------------------------------To solve this problem, my computers where overclocked (@ safe level) to accelerate the throughput and cache reduced back to 2 days. The question arose as why caching? Simple: once the rain is over, I'll still have some work left in order to reach GOLD ! MANY thanks guys! [Edit 1 times, last edit by bicotz at Dec 24, 2010 1:42:03 AM] |
||
|
anhhai
Veteran Cruncher Joined: Mar 22, 2005 Post Count: 839 Status: Offline Project Badges: |
bicotz, getting those repair jobs (the 2-3 day WUs) didn't really affected your ability to cache. Based on your description, here is what happened. And you may see it happen again soon. Every WU has an estimated time to complete set by the WCG server. Your BOINC program re-adjust that value once you get it based on your CPU and past performace, it is the value you see under the "To Completion" tab. This adjustment is re-done everytime you finish a WU, and this may cause BOINC to think that you have more or less work then you actually have. Also note, that BOINC adjusts the estimated completion time up faster then down.
----------------------------------------When I say adjust, this is what I mean. Say your cache is set to 5 days, your system can only run 1 thread at a time and you have 20 WUs all have an estimated completion time of 6 hrs. Then you do 1 WU, and it only took you 4 hrs. You will noticed that the estimated completion time of all the other WUs go down a little bit, say to 5hrs 55 mins. This let BOINC gets more WU (basically what happened when you over-clocked your system or just happen to get a fast WU). Now say after a while all of your WUs are estimated to complete around 4 hrs and then you complete another WU, but this one takes 6 hrs. Well, all of the remaining WUs estimated completion time will go up, probably close to 5.5 hrs. This cause your cache to be over full, you won't get anymore WUs until it goes down. (Please note all of this talk about estimated completion time is only the value that is displayed by the boinc manager under the "to completion" tab. BOINC then factors in the percentage of time it is on in the background) |
||
|
bicotz
Advanced Cruncher Canada Joined: Apr 25, 2010 Post Count: 67 Status: Offline Project Badges: |
It does indeed make a lot of sense. So it's a good thing to keep the cache not too full and to not play around too much with the parameters... To top it off, blend in a few sporadical overclocking and the system easilly overestimates/underestimates.
----------------------------------------To be frank, I've never been confronted with this problem for one simple reason: my cache was always set below 2 days before last weeks testings. I've marked the thread as "solved" |
||
|
Jack007
Master Cruncher CANADA Joined: Feb 25, 2005 Post Count: 1604 Status: Offline Project Badges: |
yep i hear ya bicotz,
----------------------------------------I'm keeping 4 days cache on Dengue to get to sapphire. (I might need a couple more days but I hate gettin wierd messages) As I get closer to my goal (sapphire) I'll bring it back to .1 day cache. This project has been longest to reach sapphire, I think I'm going to celebrate when I get there! GOOD LUCK on your quest for GOLD! |
||
|
anhhai
Veteran Cruncher Joined: Mar 22, 2005 Post Count: 839 Status: Offline Project Badges: |
Jack, don't lower your cache to .1 days. Leave it around 1 day unless there is some reason for the .1 day caching. The cache will help protect you against WCG going down temporarily.
---------------------------------------- |
||
|
|