| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: BOINC Agent Support
Thread: BOINC 6.10 Alpha Testing |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 70
|
|
| Author |
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Here's what I'd consider a improvement required on the 6.10 Tasks View interface. So I'm sitting there looking at the view that filters out active tasks. That view was created because if many tasks are in queue, the BM performance is sluggish and some in fact reporting it could get as bad as the BM bombing. Then when changing the view back to 'Show All Tasks", see a bunch of tasks in Uploading status. Think that "Show Active Tasks" button needs expansion to include anything that Uploading and Ready To Report maybe too. Certainly Uploading is an Activity and would signal a need to possibly check the connectivity or the message log.
----------------------------------------Another 2 cents. PS: Ingleside, 76,000 I see in BOINCview. Meantime the LTD keeps declining even if there is no work and work fetch is blocked, 6.10.10. Did that and it's crawled down now in the past 4 days to 63,000 with an STD of 0. Actually much faster that the relative weight would warrant... 520 seconds per day is it's right per day. with 14 minute jobs, that 1 every 2 days, not 10 and more every day. No, not going to debate how, why, where, when, because. This client and the previous 6.6 overfetches work and overgrants CPU time to small projects weight projects, so by when it's zero again, I might allow it to fetch a job again... thoroughly micromanaged in their case, until I get bored and leave them off completely. Detached.
WCG
----------------------------------------Please help to make the Forums an enjoyable experience for All! [Edit 1 times, last edit by Sekerob at Oct 3, 2009 9:24:30 AM] |
||
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges: 
|
PS: Ingleside, 76,000 I see in BOINCview. Meantime the LTD keeps declining even if there is no work and work fetch is blocked, 6.10.10. Did that and it's crawled down now in the past 4 days to 63,000 with an STD of 0. Actually much faster that the relative weight would warrant... 520 seconds per day is it's right per day. with 14 minute jobs, that 1 every 2 days, not 10 and more every day. No, not going to debate how, why, where, when, because. This client and the previous 6.6 overfetches work and overgrants CPU time to small projects weight projects, so by when it's zero again, I might allow it to fetch a job again... thoroughly micromanaged in their case, until I get bored and leave them off completely. Detached. It seems strange it's changing if set to "no new work"... A quick look on my system, there all projects except one is set to "no new work", gives this: 03.10.2009 13:54:09 AQUA@home [debt] CPU ineligible; debt 0.00 For all except CPDN, that was not disabled at the time, this is the exact same LTD as in my post 4 days ago, so no idea why your LTD is changing if it's been set to "no new work" the whole time... As for "overfetches", BOINC has never followed the resource-shares short-term, but tries to balance it the long term. For v6.10.xx it seems it initially wants to download upto cache-size the short-term, but after having done this initial full cache-size, it should be a very long time until next time asks for work. For the long-term, it doesn't matter if BOINC downloads and crunches example 96 hours of work on day-1, if it waits 199 days until it asks for more work. It still means 0.5% resource-share on a quad over the 200 days. This is the same result you'll get by running 0.48 hours/day on a quad, or two 14.4 minute-jobs per day.  "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." [Edit 1 times, last edit by Ingleside at Oct 3, 2009 2:40:34 PM] |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Client in the 6.10.13 iteration continues to be easily confused. Rise jobs listed at 18 hours in the buffer, rDCF 3.08. Not a specific 6.10 problem, just I'd say a kludge in the logic that is way way way overdue to be unkludged. The realities of today are multi-project DC's which can't possibly have accurate completion time predictions, making longer scheduling for off line crunchers with intermittent connectivity, days if not a week apart not a small sinecure... Now I wonder how long multiproject DC's have been running on BOINC to justify why that trac ticket remains undetermined ;>)
----------------------------------------have a nice day. PS, deep in debt at 0.59% resource share. Continuing to fetch work... maybe because the main project having 99.41% of active weight having buffered 3 times too much 09/10/2009 12:24:26 [wfd] CPU: fetch share 0.00 debt -80492.98 backoff dt 0.00 int 0.00 Their work fetch was suspended now, manually, again. edit: seems of interest... additional buffer exactly 1.00, connect 1.0 (for that 15 minute daily schedule contact that's being permitted).
WCG
----------------------------------------Please help to make the Forums an enjoyable experience for All! [Edit 1 times, last edit by Sekerob at Oct 9, 2009 11:31:00 AM] |
||
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges:
|
Now I wonder how long multiproject DC's have been running on BOINC to justify why that trac ticket remains undetermined ;>) Well, some things takes longer to implement than others... Hopefully it'll be added in v6.12.xx, but don't remember anything definite being mentioned about this at the moment. PS, deep in debt at 0.59% resource share. Continuing to fetch work... maybe because the main project having 99.41% of active weight having buffered 3 times too much 09/10/2009 12:24:26 [wfd] CPU: fetch share 0.00 debt -80492.98 backoff dt 0.00 int 0.00 Their work fetch was suspended now, manually, again. Well, you've not said what your cache-size is set at, but if it's 1 day or larger, you've still not let BOINC-client earn enough debt to stop work-request... Anyway, for v6.10.13, I've not seen any abnormalities yet... "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." |
||
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges:
|
Here's what I'd consider a improvement required on the 6.10 Tasks View interface. So I'm sitting there looking at the view that filters out active tasks. That view was created because if many tasks are in queue, the BM performance is sluggish and some in fact reporting it could get as bad as the BM bombing. Then when changing the view back to 'Show All Tasks", see a bunch of tasks in Uploading status. Think that "Show Active Tasks" button needs expansion to include anything that Uploading and Ready To Report maybe too. Certainly Uploading is an Activity and would signal a need to possibly check the connectivity or the message log. Just including "Ready to Report" will be a bad idea, since a high-end GPU-system can easily have hundreds of tasks waiting to report without there being any problems at all... But, in essence a good idea, so a variant will be added to a later client. If this will be v6.10.xx or v6.12.xx is anyone's guess, but it's atleast unlikely it will make it in initial v6.10.xx-release... Anyway, another change already made that should be included in v6.10.14 is, a re-name of "write to disk..." to "Tasks checkpoint to disk at most every N seconds", and removing the "multiply with #cpu".  "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
[ot] BOINC Tasks by Fred has an excellent implementation of view filters, summarizing Ready to Start, Ready To Report, Waiting to Run and I'm sure he will come up with others on his fast track development path... so I'll propose to him "Uploading" as an other one. What you get is a combined view of all clients, all running sciences, all tasks, so I see it just totaling 'ready to start 4 NRW, 9 HCC, 6 HCMD2 [/ot]
----------------------------------------Yes, Write to disk setting should always have been ** human logic when entering e.g. 300 seconds. Equal for all running sciences and not multiplied by the cores running on a client. So this poor guy on his octo wondering why the checkpoints are not stored in under 40 minutes. [ot2]And a snark to address one old snark v.v. large number of files ul/dl by WCG. Those GPU projects that produce hundreds of results should size up their tasks. WCG is always diligent to maintain a balance of not too long and not too short, on average. [/ot2] BTW, I would have fallen off my chair but for the arm-rests preventing that when I saw that 6.10.13 was tagged as a candidate beta... whilst 6.10.14 is already in the pipe... sigh. edit: ** read "should have remained"
WCG
----------------------------------------Please help to make the Forums an enjoyable experience for All! [Edit 1 times, last edit by Sekerob at Oct 13, 2009 8:45:23 AM] |
||
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges:
|
BTW, I would have fallen off my chair but for the arm-rests preventing that when I saw that 6.10.13 was tagged as a candidate beta... whilst 6.10.14 is already in the pipe... sigh. BOINC is continuously adding new features, improvements and bug-fixes, so can't wait until there's nothing "in the pipe", since this would mean no client would ever be released... So, if v6.10.13 passes alpha-testing, I don't see any big reason to postpone the release. "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Well Heck It's back, 6.10.13, which absolutely should NOT be released as a beta since bugs are known, reported and ignored. The wallclock/elapsed reset showed up again after a few weeks of silence. Quad concurrently running CMD2, RICE and HCC, old hands at the job. CPU run time 5:28 hours for a RICE job, 48 minutes Elapsed and 15 minutes to go at 78% progress... really, we are positive they do ~7:01 CPU hours. Berkeley needs working on their sums as this heavily affects the cache.
----------------------------------------The message log is clean, nothing but checkpoints for the past 3 hours and a gazillion debug flags. Glad I crunch with 15 minutes scheduled daily contact else my buffer would have again been in overdrive. The rain in Spain falls mainly in the week of October 22 on Barcelona, I hope. They need water anyhow and allot to iron out. NOT general populous ready, no sir. PS. It must not have happened round about when jobs finished. 2 are listed with 16:52 minutes elapsed whilst both show over 3 hours CPU time. Possibly at a checkpoint save moment BOINC goes numb in the housekeeping area? We'll keep poking. edit: Sat. 17th, 10:15 AM. It's not a bug, it's a feature! So had a bunch of jobs sitting completed, waiting to upload facing the tasks view, remember I crunch net-work suspend, but for a daily scheduled 15 minutes, and opened a second BM and went to the project tab to hit update to force a line opening for 5 minutes. In front of my eyes Elapsed time for all 4 jobs reset to zero. Reported at Berkeley, Repeated at Berkeley, Ignored at Berkeley. Since now a job finished in 8:16 minutes, though CPU time was 7 hours for a RICE, the cache was flooded once again of my supercomputer... rDCF 0.6747 17/10/2009 10:05:13 WCG update requested by user 17/10/2009 10:05:13 Resuming network activity 17/10/2009 10:05:13 WCG [sched_op_debug] Starting scheduler request 17/10/2009 10:05:14 WCG Sending scheduler request: Requested by user. 17/10/2009 10:05:14 WCG Requesting new tasks 17/10/2009 10:05:14 WCG [sched_op_debug] CPU work request: 227595.61 seconds; 0 idle CPUs 17/10/2009 10:05:14 WCG Started upload of R00364_905432d1e53f9d78b46a2060abf41404_02_011_4_0 17/10/2009 10:05:19 WCG Scheduler request completed: got 15 new tasks Saw there wont be a 6.10.13 Beta, instead a 6.10.14 alpha was released yesterday. Will it have been fixed? Not a squeak in the check in list. I rest this case.
WCG
----------------------------------------Please help to make the Forums an enjoyable experience for All! [Edit 2 times, last edit by Sekerob at Oct 17, 2009 8:22:32 AM] |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
For those in high hopes having Domain Controllers and being held down to release 5.10.45 or earlier... bummer. The decision seemingly was made by the developers to not do the installer change in 6.10., this version essentially limited to adding the possibility to participate on projects that use ATI GPU's for crunching. (counted on a few fingers).
----------------------------------------Message 28073 - Posted 17 Oct 2009 12:07:58 UTC - in response to Message 28072. The new installer use has been postponed again. It won't be in BOINC 6.10, which will instead have ATI GPU detection support. BOINC 5.10.45 can be downloaded from here (just scroll down the list).
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
hmmm, well the elapsed time resetting is back is whole or partial, the really weird part. It definitely relates to the service starting before logging in. It does not happen when the service is manually stopped and then started again using the BM. Startup delay is set to 1 minute, but can't think that could be related... not going to try and find out either as every boot takes about 5 minutes to get back to an operating state.
----------------------------------------Oh yes, it was promised to be carried to the developers before they rushed off to Spain punching out 6.10.16. Will it be fixed for the beta and 'recommended' release. PS, proposed that the developers set up a 'forum down thread' at a reliable project... their forum is down again for quite a few hours so can't tell them :-|
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
|