long standing work unit issue

Questions and Answers : Windows : long standing work unit issue
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile chthonic
Avatar

Send message
Joined: 2 Aug 99
Posts: 26
Credit: 21,795,471
RAC: 30
United States
Message 1381026 - Posted: 14 Jun 2013, 11:44:02 UTC

I stopped using seti@home awhile ago because it persistently kept ignoring files with the shortest deadline and favored files that weren't due for 1month or more. I reported this problem and it was never addressed. I am back on seti@home with a new high end machine and I see that the same issue still exists.

work units with the shorter deadlines should have priority over a file that isnt due for 2+ weeks etc... and this problem occurs regardless of my sorting by deadline in the application.. it still ignores groups of units and processes files completely out of order.

this is a waste of resources.. and should be fixed

the way I see it, units should be processed according to 3 factors (in order): Deadline, Progress and GPU (if any)

the files that require GPU resources should only run one at a time in queue and an option to auto activate GPU resources based ONLY on the presence of a unit that requires those resources.. then turns itself off if no such file exists in the queue. This way the app is not sucking up resources that it doesn't need at that point in time.

additionally.. it would be nice to tell it to focus (in order) on single files so they complete much faster.. with the extra option to tell it to focus on priority files such as GPU and Astropulse (still based on queue order) and an option to tell it how many units to queue at a time.

this way I can allocate more computing resources to the application because it wont be sucking up resources needed by other applications... it would only request those resources "as they are needed"... this is more efficient.

it would also be nice if I could tap into my file box multi-core cpu on my network without having to install a separate copy of boinc.. this box sits idle 95% of the time on my network and that idle processing power could be used for boinc on the active machine.
I treat everyone with equal disregard, I make fewer enemies that way!
ID: 1381026 · Report as offensive
Profile BilBg
Volunteer tester
Avatar

Send message
Joined: 27 May 07
Posts: 3720
Credit: 9,385,827
RAC: 0
Bulgaria
Message 1381033 - Posted: 14 Jun 2013, 12:15:59 UTC - in response to Message 1381026.  

I stopped using seti@home awhile ago because it persistently kept ignoring files with the shortest deadline and favored files that weren't due for 1 month or more. I reported this problem and it was never addressed. I am back on seti@home with a new high end machine and I see that the same issue still exists.

This is not issue, it is intended behavior.

http://boincfaq.mundayweb.com/index.php?language=1&view=168

http://boincfaq.mundayweb.com/index.php?language=1&view=541


You can artificially force (at your risk) what you want by:
Switch between tasks every ZZZZ minutes
http://setiathome.berkeley.edu/prefs.php?subset=global

Play with values in range 5000-15000 minutes
Don't go too high (e.g. >20000) or you may not get new tasks.


 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 
ID: 1381033 · Report as offensive
Profile chthonic
Avatar

Send message
Joined: 2 Aug 99
Posts: 26
Credit: 21,795,471
RAC: 30
United States
Message 1381035 - Posted: 14 Jun 2013, 12:37:42 UTC - in response to Message 1381033.  

I am aware of those postings. and the application is still "ignoring" those guidelines.
I treat everyone with equal disregard, I make fewer enemies that way!
ID: 1381035 · Report as offensive
Profile BilBg
Volunteer tester
Avatar

Send message
Joined: 27 May 07
Posts: 3720
Credit: 9,385,827
RAC: 0
Bulgaria
Message 1381040 - Posted: 14 Jun 2013, 12:54:36 UTC - in response to Message 1381035.  


What do you mean by "ignoring" those guidelines?
Exactly which 'guidelines' are ignored?

http://boinc.berkeley.edu/trac/wiki/ClientSched


 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 
ID: 1381040 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22158
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1381091 - Posted: 14 Jun 2013, 14:52:21 UTC

Generally S@H operates a FIFO scheduler for each processor class (CPU, GPU). Tasks are not shifted between their designated processor class. This works surprising well human interferance. It is known that certain other projects "bully in" with stupidly short return deadlines and so their tasks get run at high (BOINC) priority. There ways of managing these projects, but that's a different subject.
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1381091 · Report as offensive
Profile Jord
Volunteer tester
Avatar

Send message
Joined: 9 Jun 99
Posts: 15184
Credit: 4,362,181
RAC: 3
Netherlands
Message 1381098 - Posted: 14 Jun 2013, 14:57:20 UTC - in response to Message 1381026.  

work units with the shorter deadlines should have priority over a file that isnt due for 2+ weeks etc...

Let's assume you have three projects, all with their own tasks.
Project A has 14 day deadlines and tasks that run at average for 24 hours.
Project B has 28 day deadlines and tasks that run at average for 12 hours.
Project C has 7 day deadlines and tasks that run at average for 2 hours.

All projects have the same resource share, thus are weighed the same.
The computer is a quad core (4 processors), no GPU.
You're telling BOINC to cache work for minimum 2 days + 1 additional day.

Project A has 16 tasks.
Project B has 32 tasks.
Project C has 160 tasks.

If we're running strictly by deadline here's what happens:
In the above situation, only project C will run, until it manages to run all the 160 tasks to completion. When it gets below the minimum days amount, it'll ask new work, and start running that.

When will projects A and B run? Only when it's too late for them to return all of their tasks by deadline.

Thus why BOINC runs work based on first in, first out. Whatever task gets in first, be it one with a large deadline, or a small deadline, will run first.. Unless a task has a chance that it will not be run by its deadline, then it will be run in high priority.

It's also possible that when you have a lot of work in cache, that tasks run for a bit to almost done, before being stopped so other work can run, before being stopped, and yet again other work will run. it'll do this to make sure it can run all that work by deadline. BOINC calculates every second whether or not all the work in cache can be done by deadline. It does a spiffing job about it, if well left alone.
ID: 1381098 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 1381241 - Posted: 14 Jun 2013, 20:39:38 UTC - in response to Message 1381026.  

the way I see it, units should be processed according to 3 factors (in order): Deadline, Progress and GPU (if any)

The only factor that is important is that work is returned before it's deadline expires.
If the way that is accomplished upsets you, then better not to participate.
Grant
Darwin NT
ID: 1381241 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 1381243 - Posted: 14 Jun 2013, 20:41:33 UTC - in response to Message 1381098.  

BOINC calculates every second whether or not all the work in cache can be done by deadline. It does a spiffing job about it, if well left alone.

And apart from the odd issue with major BOINC manager revisions it works & works well. It's only when people start micro-managing it that things turn ugly.
Grant
Darwin NT
ID: 1381243 · Report as offensive
Profile chthonic
Avatar

Send message
Joined: 2 Aug 99
Posts: 26
Credit: 21,795,471
RAC: 30
United States
Message 1381420 - Posted: 15 Jun 2013, 6:23:44 UTC

I wont accept that this can't be done until an official staffer confirms beyond all doubt that it isn't possible.
I treat everyone with equal disregard, I make fewer enemies that way!
ID: 1381420 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 1381426 - Posted: 15 Jun 2013, 6:52:38 UTC - in response to Message 1381420.  

I wont accept that this can't be done until an official staffer confirms beyond all doubt that it isn't possible.

No one has said it's not possible, just that there is no reason to.
Work is done in the order it is received, unless there is a risk of it missing a deadline.
Just because someone would like work to be processed in a particular order isn't reason enough to spend the time & effort necessary to make it possible.

Maybe when they've done all the work that needs to be done, and all the other things that would make things easier to use, and 1,001 other things that people would like, then maybe allowing people to configure the order work is processed might make it on their list of things to do. But i doubt it as i would expect you'd still get upset when it processes work in a different order to what you would like because something was at risk of missing it's deadline.
Grant
Darwin NT
ID: 1381426 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 1381428 - Posted: 15 Jun 2013, 6:58:31 UTC - in response to Message 1381026.  
Last modified: 15 Jun 2013, 7:07:45 UTC

it would also be nice if I could tap into my file box multi-core cpu on my network without having to install a separate copy of boinc.. this box sits idle 95% of the time on my network and that idle processing power could be used for boinc on the active machine.


Why on earth would you want to do that?
The advantage of having the manager on the machine doing the work is it is resposnible for just *that* machine, and you don't have huge amounts of network traffic between that machine & the one(s) that has the manager on it.
You can always remotely control BOINC installations on remote machines- that's how people with 100s of systems do it.

EDIT what you are describing is clustering & is a completely different way of processing work, not what Seti is based on (or any of the other BOINC projects- that's why they are using the BOINC distributed computing framework. If they had the funds to use clusters, they would use them).


As for the rest of your wish list- i honestly can't work out what it is you want.
Grant
Darwin NT
ID: 1381428 · Report as offensive

Questions and Answers : Windows : long standing work unit issue


 
©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.