5.8.3 Time to Completion & RDCF

Message boards : BOINC Manager : 5.8.3 Time to Completion & RDCF
Message board moderation

To post messages, you must log in.

AuthorMessage
W-K ID 666

Send message
Joined: 30 Dec 05
Posts: 465
United Kingdom
Message 7665 - Posted: 21 Jan 2007, 3:05:44 UTC

Ive been running 5.8.x Since mid December and 5.8.3 for the last four days. And have notice a problem with 5.8.3 and the Time to Completion (TtC) and the Result Duration Correction Factor, (RDCF). This is on my Pent M computer.
At first I thought the TtC had been calculated without taking the RDCF into account, and informed JM7.
On closer inspection, I noticed, for each Project, that the TtC increased each time a unit completed. And therefore checked and recorded my RDCF for each project. The RDCFs for each project are going up slowly. But all units are completing much faster than the original TtC estimate.

The RDCF should increase sharply, when a unit exceeds the TtC and decrease by 10% of the difference between estimate and actual. As quote by JM7 and Dr.A in their paper on BOINC Client Scheduling;
BOINC corrects for these factors with a per-project
duration correction factor, an estimate of the ratio of
actual CPU time to originally estimated CPU time. As
will be seen, it is better to err on the high side, so this is
calculated in an asymmetric way; increases are
reflected immediately, but decreases are exponentially
smoothed.

I've not had the time, to monitor this situation closely but it appears that the initial TtC, and therefore the RDCF, is being increased rather than decreased by result the formula ('Estimated time' - 'actual time') * 0.1.

If anybody else is seeing this could they please post here or info JM7 directly.

Andy
ID: 7665 · Report as offensive
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15512
Netherlands
Message 7666 - Posted: 21 Jan 2007, 3:16:16 UTC

Lots of other people see this in the latter parts of this thread. I already asked JM7 about it and only got as answer he didn't change the DCF code.

But I am glad you bring it back up, Andy.
ID: 7666 · Report as offensive
W-K ID 666

Send message
Joined: 30 Dec 05
Posts: 465
United Kingdom
Message 7675 - Posted: 21 Jan 2007, 6:38:56 UTC

I got similar replies from JM7, He seems to think that I didn't understand the varying lengths of unit on Seti. This computers ID on Beta is 351, and my observation to him was about four consecutive units on Beta from the same batch with identical initial TtC, and I caught it, by chance as the 2nd of these was just finishing. ANd observed the TtC increase when it should have decreased.

My other computer is a C2D, well my 20something sons actually, its running 5.8.0 at moment, I will later today upgrade that today. Once he has woken and a path can be cleared from door to computer. And then will monitor RDCF from here by visiting the website. It's easier that trying to do it in person or by controlling it as I never know when BOINC is running. Last time I tried it I crashed a network game he had been running for several hours, I was not his favourite person for days.

Andy
ID: 7675 · Report as offensive
W-K ID 666

Send message
Joined: 30 Dec 05
Posts: 465
United Kingdom
Message 8063 - Posted: 8 Feb 2007, 2:56:33 UTC

Having now upgraded to BOINC 5.8.9, it appears the problem with RDCF, and associated times is now fixed.

Thank you Dr.A, JM7 and any others for your work.

Andy
ID: 8063 · Report as offensive
W-K ID 666

Send message
Joined: 30 Dec 05
Posts: 465
United Kingdom
Message 8089 - Posted: 9 Feb 2007, 3:42:36 UTC
Last modified: 9 Feb 2007, 3:43:41 UTC

Update on this now my C2D has reported 5+ units on each of Seti, SetiBeta and Einstein.

The RDCF has indeed gone down, but I'm wondering if it isn't now going down too far or too fast. These are approx figures for older versions and the figs as I type for 5.8.9
		5.4.11	5.8.0/8	5.8.9
Einstein	~0.55	>0.7	0.5078
Seti		<0.5	~0.6	0.337394
SetiBeta	<0.6	>0.65	0.468397

Andy
ID: 8089 · Report as offensive
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15512
Netherlands
Message 8101 - Posted: 9 Feb 2007, 13:14:26 UTC

In my experience, they'll go a little under their minimum (so the estimates are less than actual run time), before leveling off at the right time(s).
ID: 8101 · Report as offensive

Message boards : BOINC Manager : 5.8.3 Time to Completion & RDCF

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.