Message boards :
Number crunching :
Merry Christmas and a Happy New Year !!
Message board moderation
Author | Message |
---|---|
Keith T. Send message Joined: 23 Aug 99 Posts: 962 Credit: 537,293 RAC: 9 |
My Athlon XP has recently completed 4 WUs with a deadline of 25th Dec '07. It's just reported one with a 31 December deadline. I think others have already seen WUs with 2008 deadlines. Do we really need deadlines this long ? Meanwhile my old P233 MMX is crunching away on a Christmas WU which should be done by the middle of this week. It's already over 57.5 % done after just over 75 hours. It's still got a pending that took ~ 8 days to crunch, that was reported on 24th October. That one will time out on my birthday, 11th December. I know a few are still using hardware slower than a P233 MMX, but there can't be many. Just for fun I may install SETI on an old P100 laptop that is in the back of a cupboard, I don't think it would have too many problems with the current deadlines ! Sir Arthur C Clarke 1917-2008 |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14653 Credit: 200,643,578 RAC: 874 |
And a happy Valentine's Day to you! Edit: Looks like the wingman might finally reach it today, so for posterity: 648964319 3751792 1 Nov 2007 3:22:49 UTC 2 Nov 2007 8:05:53 UTC Over Success Done 9,828.84 111.14 pending 648964320 3221936 1 Nov 2007 7:25:53 UTC 14 Feb 2008 1:06:56 UTC In Progress Unknown New --- --- --- |
Keith T. Send message Joined: 23 Aug 99 Posts: 962 Credit: 537,293 RAC: 9 |
Potential "wingmen" can breathe a sigh of releif. The P100 booted OK (last switched on in April 2006), but it only has 16 MB RAM. P233 has now passed 59% on it's Xmas WU and should finish around the 8th of this month according to BoincView estimate. |
Keith T. Send message Joined: 23 Aug 99 Posts: 962 Credit: 537,293 RAC: 9 |
And a happy Valentine's Day to you! Thanks Richard, I thought I has seen some reports of 2008 deadlines appearing already. Has anyone got one later than this yet ? Sir Arthur C Clarke 1917-2008 |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Long deadlines make use of PCs with only small % of CPU time dedicated to BOINC. So it's not nessesary old and slow CPUs who may miss deadline, shouldn't forget about this ;) So as long as server database can afford increase in size to hold many unfinished WUs long deadlines are good IMHO :) |
Josef W. Segur Send message Joined: 30 Oct 99 Posts: 4504 Credit: 1,414,761 RAC: 0 |
Long deadlines make use of PCs with only small % of CPU time dedicated to BOINC. So it's not nessesary old and slow CPUs who may miss deadline, shouldn't forget about this ;) I agree. The deadline is intended to represent what can be accomplished with a machine having a Whetstone MIPS rating of 33.3 and crunching S@H 24/7. That would also match: 100 WMIPS crunching 3 projects 24/7. 1000 WMIPS crunching 30 projects 24/7. 3000 WMIPS crunching 30 projects an average of 8 hours per day. 2000 WMIPS crunching 2 projects an average of 24 minutes per day. Joe |
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
My Athlon XP has recently completed 4 WUs with a deadline of 25th Dec '07. It's just reported one with a 31 December deadline. I think others have already seen WUs with 2008 deadlines. Do we really need deadlines this long? I'd turn this around and say "do we really need people obsessing about EDF?" Because, if the deadlines were shorter, BOINC would go into "Earliest Deadline First" mode, and process work based on making deadlines -- and accumulating debt that has to be balanced out later. If you go back before Multibeam, you'll see posts where people are crying because BOINC won't download more SETI, and is working on other projects exclusively. Yes, crying. Complaining bitterly that BOINC is badly broken, that it couldn't schedule its way out of a paper bag. Fiddling with settings to make BOINC do more SETI, even though BOINC is trying valiantly to return work on time and meet resource shares. ... even though, left alone, BOINC will do what it has been asked to do. Long deadlines relieve the schedule pressure, and relaxed schedules make EDF very unlikely. You can't have it both ways -- people need to recognize that EDF is your friend, or they need to stop worrying about the long deadlines. -- Ned |
Brian Silvers Send message Joined: 11 Jun 99 Posts: 1681 Credit: 492,052 RAC: 0 |
The only problem, Joe, is that the bulk of the machines causing very lengthy pending times, and thus the "why do I have so much pending" questions from other users, ARE NOT these slower machines... They appear to be faster machines with people building up large caches and then taking a very long time to get around to working on them, or people that grab results and then decide to quit, for whatever reason. Looking at the 2+ week pendings on my hosts, I have the following cpu types:
|
Keith T. Send message Joined: 23 Aug 99 Posts: 962 Credit: 537,293 RAC: 9 |
Long deadlines make use of PCs with only small % of CPU time dedicated to BOINC. So it's not nessesary old and slow CPUs who may miss deadline, shouldn't forget about this ;) My P233 is currently attached to 4 projects: SETI Resource Share 50 SETI Beta Resource Share 10 LHC@home Resource Share 25 Leiden Classical Resource Share 2.5 It currently has work for 3 of these, closest deadline is tomorrow but it should make it without problems as that WU is currently 99.303% complete and Waiting to run. It's other 2 WUs are both due on Christmas Day. SETI main is now up to 64.3% complete after ~ 81.5 hours. SETI Beta has only done 4.75% @ just under 8 hours so far. BOINC version was recently updated to 5.10.28 [EDIT] I have resisted the temptation to fiddle with BOINC buttons to get that Leiden WU complete, its currently @ -25,185 Short Term Debt, Waiting to run, according to BoincView, but time to complete is estimated @ 12 mins 58 seconds. [/EDIT] Sir Arthur C Clarke 1917-2008 |
Brian Silvers Send message Joined: 11 Jun 99 Posts: 1681 Credit: 492,052 RAC: 0 |
The deadlines might be a bit too long. I personally think they could be shortened by 2 weeks for the longest results and still not cause serious harm (what you termed as "crying")... What I see time and time again when looking at people that post about pending credit issues is that they are getting paired up with some people who have highly capable machines that appear to have just went out in la-la for whatever reason. It would be interesting to find out if there is a stat somewhere that could tell us how many processors there are that are slower than a 233 MMX. Toby's site may have it... Dunno... |
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
Brian, my friend, my point isn't that EDF is a problem. It's the crying that is the problem. The deadlines could be shortened significantly more than 2 weeks for the longest work units. I have a kind-of mid-range machine, and it is predicting 9 hours for work due in late December. Cut the time to two weeks, and it'll still finish comfortably. What it won't do is work on each one in the order downloaded, it will instead do them in deadline order. It may shift to doing SETI Beta exclusively when I want it to do SETI. But, if I leave it well enough alone, it'll do "too much" SETI Beta, and then it won't do any for a while, and everything balances out. I've studied it, and it all works just fine. EDF is your friend, especially on glacially slow machines like my 800 MHz C3. ... but people clearly don't like EDF, so the project gave us deadlines that avoid EDF -- and now the complaints are "deadlines way too long." The only part that has me mildly bothered is the fact that people seem too easily upset by what seems to be improper behaviour at first glance, when BOINC really handles all of these conflicting requirements very well. There is no "serious harm" from running in EDF 100% of the time. The crying is based on perception, not reality. -- Ned |
Brian Silvers Send message Joined: 11 Jun 99 Posts: 1681 Credit: 492,052 RAC: 0 |
I understood... :)
...and to prevent said crying, that's why I suggested only shortening a small amount. Two weeks for the longer tasks would put a burden on anyone who has less than a 1GHz machine and is active with Einstein. Hell, my machine that's an FX-57 equivalent takes 10-13 hours of CPU time for a single Einstein task right now. Alinator's K6 300 would likely push the 2-week limit as it is for just one of the S5R3 units there... We'll see how the K6-2 does, as he has one running (I think). Edit: I looked at Tullio's host and it took 5.5 days of CPU time for an Einstein S5R3. However, it being an Intel processor, it has a better FPU than a K6-2. You'd probably be OK with a bottom-end Athlon (500MHz), but even a K6-2 may struggle to meet a two week deadline for both projects...
They are way too long, and it's not totally about pending credit. I'm seeing hosts pick up hundreds, even thousands, of results that have no chance of completing. The project may as well set IR back up to 3 and issue aborts for the trailer if it hasn't started. That will have the extra bonus of helping to reduce the pre-5.2.6 clients' impact as well... Brian |
popandbob Send message Joined: 19 Mar 05 Posts: 551 Credit: 4,673,015 RAC: 0 |
Maybe it needs to be implemented that a certain CPU can only have a max number of wu's (could be done via benchmark speed also?) and also re-issue results that are on fast machines that are not returning work after a few weeks. But then do we have the servers to be able to handle the extra work? ~BoB Do you Good Search for Seti@Home? http://www.goodsearch.com/?charityid=888957 Or Good Shop? http://www.goodshop.com/?charityid=888957 |
Scarecrow Send message Joined: 15 Jul 00 Posts: 4520 Credit: 486,601 RAC: 0 |
Merry Christmas and a Happy New Year !! Thank [PoliticallyCorrectMode] your favorite deity [/PoliticallyCorrectMode] you didn't suggest a gift exchange. |
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
The idea here is that you'd get one work unit for one project, go into EDF and the computer would do just that one work unit at least until it was obvious that deadlines would be made, then do the other work. Long Term Debt "gates" the downloads so that SETI would stop downloading until Einstein was caught up, or the reverse. Tullio's machine, at 296 FLOP/s is almost twice as fast as my C3 800 MHz. The FPU on the C3 is not quite nonexistent, but it's dismally slow -- 170 million FLOP/s. The C3 does a .43 angle range WU (54 credits) in about 55 hours. It will return work sometime today that was downloaded on the 1st. It has a 10.25 day average turn-around. If the deadline on this work unit was November 18th instead of December 18th, it'd still be returned with plenty of time to spare. ... and if the slug crunched for multiple projects, it'd pretty much keep one WU in the cache, with the next one downloaded a couple of hours before it ran out. |
Brian Silvers Send message Joined: 11 Jun 99 Posts: 1681 Credit: 492,052 RAC: 0 |
OK, but I was under the (perhaps mistaken) assumption that SETI does more integer math than Einstein... If you'd like to participate in an experiment, try attaching to Einstein with that and see how long a S5R3 unit takes... |
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
The initial prediction is 317 hours. |
Brian Silvers Send message Joined: 11 Jun 99 Posts: 1681 Credit: 492,052 RAC: 0 |
If you'd like to participate in an experiment, try attaching to Einstein with that and see how long a S5R3 unit takes... 317/24 = 13.21 days. Deadlines there are 14 days. Your time to completion will be rising, perhaps substantially, especially at first. You should either see steady rising or rising followed by a stepped drop, then rising again. Someone to baseline against would be Alinator. His K6 300 may or may not be active. For an S5R2 result, he estimated it would need about 800 hours. The deadlines then were 3 weeks. The S5R3 units are about 1/3 the size (roughly), but are not 1/3 of the runtime (more like 1/2)... At any rate, I know your point is that it would only work on one task, but until BOINC figured it out, it still should switch tasks... Lowering deadlines too much will make SETI too unfriendly with multiple projects attached...especially Einstein, given that it will already be tight for slower hosts... There needs to be a bit more balanced setting for the deadline here, but not cut back to an extreme. As always, YMMV... |
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
At any rate, I know your point is that it would only work on one task, but until BOINC figured it out, it still should switch tasks... In a sense, it does switch tasks -- except that it switches on a much longer time frame. My point really is that if you ask for a 50/50 share between two projects, you'll get that. It may take weeks, but you'll get a 50/50 share. |
Brian Silvers Send message Joined: 11 Jun 99 Posts: 1681 Credit: 492,052 RAC: 0 |
Ah, but with the associated EDF, which will cause crying and gnashing of fingers... I'm still curious about db expansion. The deadlines for the earliest "long deadline" units should be happening around now... |
©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.