Message boards : Number crunching : Strange deadlines
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile slavko.sk

Send message
Joined: 28 Sep 04
Posts: 15
Credit: 176,263
RAC: 0
Message 9715 - Posted: 30 Aug 2005, 8:33:47 UTC

A few days ago I donwloaded few WU's with deadline of 7th september.
But from yesterday or so I'm downloading WU's with deadline of 3rd september. If this will continue, then WU's with deadline of 7th september will expire.
Is this ok, or WU's which I'm recieving now are some "fast" to compute first and in a few days they will stop distribute them.
Does anybody know about this?
ALL GLORY TO THE HYPNOTOAD!
Potrebujete pomoc?
ID: 9715 · Report as offensive     Reply Quote
Gaspode the UnDressed

Send message
Joined: 1 Sep 04
Posts: 506
Credit: 118,619
RAC: 0
Message 9716 - Posted: 30 Aug 2005, 8:42:09 UTC - in response to Message 9715.  
Last modified: 30 Aug 2005, 8:42:49 UTC

<blockquote>A few days ago I donwloaded few WU's with deadline of 7th september.
But from yesterday or so I'm downloading WU's with deadline of 3rd september. If this will continue, then WU's with deadline of 7th september will expire.
Is this ok, or WU's which I'm recieving now are some "fast" to compute first and in a few days they will stop distribute them.
Does anybody know about this?</blockquote>

A couple of weeks ago LHC@Home installed a deadline optimiser that looks at the available computing capacity and estimates the deadline that will complete the study fastest. After they did this the deadlines started dropping - first to eight days, and now five. A shorter deadline means that WUs will be re-issued sooner if they're not returned, so the study should complete earlier.

Your machines should crunch WUs in the order they are received, unless deadlines start approaching. In this case the client should switch to Earliest Deadline First (EDF) mode, and later WUs with earlier deadlines will crunch first.

You should be fine unless you have a cache set to several days.

I suspect that the tight deadlines will drive away some participants who can't or don't wish to provide computing resources to meet the tight schedule. We'll see!


Gaspode the UnDressed
http://www.littlevale.co.uk
ID: 9716 · Report as offensive     Reply Quote
Profile slavko.sk

Send message
Joined: 28 Sep 04
Posts: 15
Credit: 176,263
RAC: 0
Message 9717 - Posted: 30 Aug 2005, 8:46:52 UTC - in response to Message 9716.  

<blockquote>Your machines should crunch WUs in the order they are received, unless deadlines start approaching. In this case the client should switch to Earliest Deadline First (EDF) mode, and later WUs with earlier deadlines will crunch first.</blockquote>

Computer is running in "29.8.2005 22:07:48||Using earliest-deadline-first scheduling because computer is overcommitted." EDF mode, so processing is right. Only question is how it will continue and if it will catch WU's with later deadline on time.
ALL GLORY TO THE HYPNOTOAD!
Potrebujete pomoc?
ID: 9717 · Report as offensive     Reply Quote
Gaspode the UnDressed

Send message
Joined: 1 Sep 04
Posts: 506
Credit: 118,619
RAC: 0
Message 9718 - Posted: 30 Aug 2005, 9:45:50 UTC - in response to Message 9717.  

<blockquote>
Computer is running in "29.8.2005 22:07:48||Using earliest-deadline-first scheduling because computer is overcommitted." EDF mode, so processing is right. Only question is how it will continue and if it will catch WU's with later deadline on time.</blockquote>

If your machine says it is overcommitted then there is a chance that it won't complete WUs by the deadline. If you're running another BOINC project then you could suspend it or reduce its resource share to give LHC a bit more CPU time, but you might miss deadlines elsewhere. Keep a close eye on it...

Gaspode the UnDressed
http://www.littlevale.co.uk
ID: 9718 · Report as offensive     Reply Quote
Profile The Gas Giant

Send message
Joined: 2 Sep 04
Posts: 309
Credit: 715,258
RAC: 0
Message 9719 - Posted: 30 Aug 2005, 10:41:36 UTC

I wouldn't worry too much. Since the wu estimation times are sooo wrong even under 4.72 which is an improvement, it means that the total work that BOINC thinks you have is actually less than what you actually have, so it should all be completed before the deadlines are reached. You should also see a message that says something like "...work fetch is suspended."

Live long and crunch.

Paul
(S@H1 8888)
BOINC/SAH BETA
ID: 9719 · Report as offensive     Reply Quote
Profile slavko.sk

Send message
Joined: 28 Sep 04
Posts: 15
Credit: 176,263
RAC: 0
Message 9720 - Posted: 30 Aug 2005, 10:44:31 UTC

Yes, I already seen massage: 30.8.2005 11:17:48|LHC@home|Message from server: (won't finish in time) Computer on 96.6% of time, BOINC on 99.2% of that, this project gets 40.0% of that
Which prevent to download new work. I will keep an eye on it, it is my notebook, no problem. Just wondering how they distribute work. In general it is nice.
ALL GLORY TO THE HYPNOTOAD!
Potrebujete pomoc?
ID: 9720 · Report as offensive     Reply Quote
Profile Chrulle

Send message
Joined: 27 Jul 04
Posts: 182
Credit: 1,880
RAC: 0
Message 9806 - Posted: 1 Sep 2005, 9:35:59 UTC

Remember also that a deadline is not a hard deadline. You can still return a result for some time after the deadline and still have a reaonable time to get it accepted by the server.

Chrulle
Research Assistant &amp; Ex-LHC@home developer
Niels Bohr Institute
ID: 9806 · Report as offensive     Reply Quote
Profile slavko.sk

Send message
Joined: 28 Sep 04
Posts: 15
Credit: 176,263
RAC: 0
Message 9813 - Posted: 1 Sep 2005, 11:12:31 UTC

Mentioned box finished all "short deadline" WU in a meantime. So everything goes well. I'm waiting for new server which I will connect also to this project. Hope, here will be enough work in hte future.
ALL GLORY TO THE HYPNOTOAD!
Potrebujete pomoc?
ID: 9813 · Report as offensive     Reply Quote

Message boards : Number crunching : Strange deadlines


©2024 CERN