What determines how much work to request?

Author	Message
adrianxw Send message Joined: 2 Oct 05 Posts: 400	Message 4375 - Posted: 18 May 2006, 10:12:57 UTC Last modified: 18 May 2006, 10:24:17 UTC I have 2 machines, one running 5.2.13, the other 5.4.9, both run a variety of projects at different quotas. Both connect every 0.1 days and are permenantly coupled to the net. With the CASP experiment starting, I have moved more resources to Rosetta recently, hence some other projects have reduced somewhat. My interest right now is with MalariaControl.net. MCDN has really rather short deadlines, typically a few days. Now I have set it's CPU quota to 6% on one of my machines. This is easily enough to crunch a wu within the deadline without going into EDF. The issue is, however, that MCDN always seems to download 4-5 wu's at a time. Thus, the machine says, whoa, overload, EDF. So I get a bunch of MCDN, then none for a couple of days while the negative LTD works off, then another bunch. Before anyone says it, yes I know EDF is not a problem, it is just the system doing what it is supposed to do. I know how LTD and EDF work, I just do not know why MCDN requests "excessively" when the other projects do not. Here are my "Requesting" lines for the last week or so, as you can see, most projects, most of the time ask for a little work, but MCDN is requesting 17280 seconds, (4.8 hours), each time. Why is that, and can I alter this behaviour? (This log is from the 5.4.9 client). Hint, look at the longest lines. 2006-05-09 01:29:31 [boincsimap] Requesting 101 seconds of new work 2006-05-09 02:30:52 [rosetta@home] Requesting 1214 seconds of new work 2006-05-09 13:18:15 [rosetta@home] Requesting 2477 seconds of new work 2006-05-09 14:07:40 [Predictor @ Home] Requesting 21 seconds of new work 2006-05-09 16:43:45 [rosetta@home] Requesting 17280 seconds of new work 2006-05-09 18:23:03 [Predictor @ Home] Requesting 56 seconds of new work 2006-05-09 21:31:55 [boincsimap] Requesting 232 seconds of new work 2006-05-10 01:41:39 [Predictor @ Home] Requesting 100 seconds of new work 2006-05-10 02:38:31 [rosetta@home] Requesting 3116 seconds of new work 2006-05-10 05:39:35 [malariacontrol.net beta] Requesting 17280 seconds of new work 2006-05-10 06:40:49 [rosetta@home] Requesting 2603 seconds of new work 2006-05-10 12:31:51 [boincsimap] Requesting 75 seconds of new work 2006-05-10 16:10:09 [rosetta@home] Requesting 4131 seconds of new work 2006-05-10 16:50:30 [Predictor @ Home] Requesting 51 seconds of new work 2006-05-10 20:52:23 [rosetta@home] Requesting 1777 seconds of new work 2006-05-10 21:30:37 [Predictor @ Home] Requesting 3460 seconds of new work 2006-05-10 21:30:45 [rosetta@home] Requesting 1893 seconds of new work 2006-05-10 22:30:43 [Predictor @ Home] Requesting 3444 seconds of new work 2006-05-10 22:50:15 [Einstein@Home] Requesting 143 seconds of new work 2006-05-11 00:41:08 [boincsimap] Requesting 16 seconds of new work 2006-05-11 05:01:51 [Predictor @ Home] Requesting 8 seconds of new work 2006-05-11 05:24:20 [rosetta@home] Requesting 872 seconds of new work 2006-05-11 07:55:10 [boincsimap] Requesting 2 seconds of new work 2006-05-11 09:38:59 [rosetta@home] Requesting 1067 seconds of new work 2006-05-11 14:09:40 [Predictor @ Home] Requesting 26 seconds of new work 2006-05-11 14:13:51 [malariacontrol.net beta] Requesting 17280 seconds of new work 2006-05-11 20:21:35 [rosetta@home] Requesting 2628 seconds of new work 2006-05-11 21:42:17 [boincsimap] Requesting 8279 seconds of new work 2006-05-11 21:42:27 [Predictor @ Home] Requesting 220 seconds of new work 2006-05-11 21:45:01 [SETI@home] Requesting 17280 seconds of new work 2006-05-12 03:27:52 [rosetta@home] Requesting 1059 seconds of new work 2006-05-12 09:28:50 [malariacontrol.net beta] Requesting 17280 seconds of new work 2006-05-12 10:26:26 [rosetta@home] Requesting 4162 seconds of new work 2006-05-12 19:33:23 [rosetta@home] Requesting 2069 seconds of new work 2006-05-12 21:28:40 [Predictor @ Home] Requesting 119 seconds of new work 2006-05-13 01:51:30 [rosetta@home] Requesting 2553 seconds of new work 2006-05-13 08:04:06 [Predictor @ Home] Requesting 51 seconds of new work 2006-05-13 09:38:05 [Predictor @ Home] Requesting 2851 seconds of new work 2006-05-13 10:33:30 [rosetta@home] Requesting 2440 seconds of new work 2006-05-13 11:35:32 [boincsimap] Requesting 21 seconds of new work 2006-05-13 13:03:58 [malariacontrol.net beta] Requesting 17280 seconds of new work 2006-05-13 18:48:56 [rosetta@home] Requesting 912 seconds of new work 2006-05-14 03:14:59 [rosetta@home] Requesting 17280 seconds of new work 2006-05-14 04:45:28 [boincsimap] Requesting 54 seconds of new work 2006-05-14 10:33:47 [rosetta@home] Requesting 17280 seconds of new work 2006-05-14 15:30:39 [rosetta@home] Requesting 46 seconds of new work 2006-05-14 18:16:30 [boincsimap] Requesting 46 seconds of new work 2006-05-14 19:48:54 [Predictor @ Home] Requesting 27 seconds of new work 2006-05-14 22:56:42 [rosetta@home] Requesting 3491 seconds of new work 2006-05-15 06:20:19 [rosetta@home] Requesting 17280 seconds of new work 2006-05-15 09:45:01 [boincsimap] Requesting 51 seconds of new work 2006-05-15 10:09:50 [Predictor @ Home] Requesting 36 seconds of new work 2006-05-15 11:10:03 [malariacontrol.net beta] Requesting 17280 seconds of new work 2006-05-15 13:30:22 [rosetta@home] Requesting 2906 seconds of new work 2006-05-15 20:33:09 [rosetta@home] Requesting 3720 seconds of new work 2006-05-16 03:48:01 [rosetta@home] Requesting 3357 seconds of new work 2006-05-16 05:24:07 [boincsimap] Requesting 88 seconds of new work 2006-05-16 06:40:32 [Predictor @ Home] Requesting 55 seconds of new work 2006-05-16 12:32:16 [rosetta@home] Requesting 4671 seconds of new work 2006-05-16 16:54:15 [Predictor @ Home] Requesting 8 seconds of new work 2006-05-16 17:54:28 [Einstein@Home] Requesting 17280 seconds of new work 2006-05-16 19:26:50 [boincsimap] Requesting 83 seconds of new work 2006-05-16 19:55:23 [rosetta@home] Requesting 1240 seconds of new work 2006-05-17 00:41:22 [rosetta@home] Requesting 2049 seconds of new work 2006-05-17 05:53:14 [malariacontrol.net beta] Requesting 17280 seconds of new work 2006-05-17 12:02:50 [rosetta@home] Requesting 1776 seconds of new work 2006-05-17 13:35:23 [boincsimap] Requesting 24 seconds of new work 2006-05-17 16:09:55 [Predictor @ Home] Requesting 17280 seconds of new work 2006-05-17 18:43:37 [rosetta@home] Requesting 3133 seconds of new work 2006-05-18 01:43:51 [rosetta@home] Requesting 3652 seconds of new work 2006-05-18 03:13:49 [boincsimap] Requesting 61 seconds of new work 2006-05-18 08:19:37 [boincsimap] Requesting 108 seconds of new work 2006-05-18 09:42:41 [rosetta@home] Requesting 1316 seconds of new work Wave upon wave of demented avengers march cheerfully out of obscurity into the dream. ID: 4375 ·

adrianxw Send message Joined: 2 Oct 05 Posts: 400	Message 4382 - Posted: 18 May 2006, 16:23:13 UTC It won't let me edit. I looked further back, and actually it DOES sometime ask for less, but very rarely, so the question stands. Wave upon wave of demented avengers march cheerfully out of obscurity into the dream. ID: 4382 ·

Keck_Komputers Send message Joined: 29 Aug 05 Posts: 304	Message 4394 - Posted: 19 May 2006, 10:24:08 UTC A request normally asks for enough work to top off the queue. Since there is no malaria work in the queue it asks for more for that project than it would for a project that already had work. BOINC WIKI BOINCing since 2002/12/8 ID: 4394 ·

Norbert Hoffmann Send message Joined: 19 Dec 05 Posts: 28	Message 4399 - Posted: 19 May 2006, 15:40:02 UTC - in response to Message 4375. Here are my "Requesting" lines for the last week or so, as you can see, most projects, most of the time ask for a little work, but MCDN is requesting 17280 seconds, (4.8 hours), each time. That's because LTD works like a switch and not like an intelligent controller. That is: after EDF MC has collected enough negativ LTD that the download scheduler decides to fetch no new work for a time. When LTD gets a bit less negativ, downloading is allowed again and a full queue is requested (leading to EDF again). Adding to the problem is, that the scheduler at the server side seems not to take the CPU quota into account when calculating how much work is needed. So it sends even more than requested. Norbert ID: 4399 ·

adrianxw Send message Joined: 2 Oct 05 Posts: 400	Message 4408 - Posted: 20 May 2006, 16:12:34 UTC So, EDF is the problem. If I let MCDN run down, then zap it's LTD, it will then behave like the other science clients and download sensibly? Wave upon wave of demented avengers march cheerfully out of obscurity into the dream. ID: 4408 ·

Norbert Hoffmann Send message Joined: 19 Dec 05 Posts: 28	Message 4409 - Posted: 20 May 2006, 17:36:15 UTC - in response to Message 4408. So, EDF is the problem. EDF is no problem. It's necessary sometimes. If I let MCDN run down, but not to 0, let it run until it's time to complete equals your queue and quota settings then zap it's LTD, better zap STD and LTD it will then behave like the other science clients yes and download sensibly? no :) Norbert ID: 4409 ·

Copyright © 2024 University of California.
Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.