Work unit shortage idea


log in

Advanced search

Questions and Answers : Wish list : Work unit shortage idea

Author Message
Darren
Volunteer tester
Avatar
Send message
Joined: 2 Jul 99
Posts: 259
Credit: 276,694
RAC: 55
United States
Message 2153 - Posted: 25 Jun 2004, 22:48:06 UTC

OK, this is not going to win me any friends I'm sure, but after spending a few days trying to help people get boinc up and running and seeing how many people think they have done something wrong because they can't get any work, I think maybe it might be justified.

During times of EXTREME work unit shortage, why not limit the number of downloads by ACCOUNT -- when the critical point is reached, stop serving work to all but the one top ranked host for each account.

In my case, I have 7 hosts, and while that certainly won't set any records I still like points just as much as the next guy. But, I would not be the slightest bit upset if Berkeley stopped feeding my 6 least productive hosts - as long as the same thing was being done to everyone else. I would simply turn them off myself, but all that would accomplish would be that the few work units they are getting would be sucked up by some IT guy running 1000 hosts that aren't even his (at least one of which, at any given moment in time, will likely be trying to get work). If Berkeley stopped feeding me with a guarantee that it would help out some of these poor little 1 computer guys/gals just trying to get one single work unit to do - so be it, I certainly wouldn't complain about it.

Now, call me a bleeding heart liberal if you must (no doubt I've been called worse), but it kind of pisses me off to see people who have gotten thousands of work units (and have hundreds still pending), while some little guy trying his damnedest to help the cause the best that he can is just getting stomped all over and can't even get one.

In the absence of some limiting feature like my suggestion above, there at least needs to be one redundant work unit that everybody starts every host with. It could give minimal (or even 0) credit, but the first time you connect a new host, that is the work unit you get. That would serve the purpose of giving the people who are interested a way to see how their computer really stacks up with others, as they could check in their account and see what other computers really did with the exact same work unit. It would also - and more importantly - give EVERYONE at least one work unit at the onset, then they will at least know that their system does, in fact, work.



Profile Shaktai
Volunteer tester
Avatar
Send message
Joined: 16 Jun 99
Posts: 211
Credit: 259,752
RAC: 0
United States
Message 2168 - Posted: 25 Jun 2004, 23:25:20 UTC

Darren,

Not a bad idea, but I think that instead of patching the system to fix a temporary problem, the project managers have decided to utilize their limited resources to fix the problem, by bringing more resources to bear on it. Just one of those decisions that has to be made when you have finite computer and human resources.

Heffed
Volunteer tester
Send message
Joined: 19 Mar 02
Posts: 1856
Credit: 40,736
RAC: 0
United States
Message 2218 - Posted: 26 Jun 2004, 2:12:54 UTC

Not a bad concept, but I'm not sure how it could be implemented.

Bad as Shaktai has mentioned, they're adding more hardware.

As mentioned on the front page, they are bringing a third work splitter online, and are rewriting a more effecient transitioner.

John McLeod VII
Volunteer developer
Volunteer tester
Avatar
Send message
Joined: 15 Jul 99
Posts: 24520
Credit: 521,562
RAC: 92
United States
Message 2267 - Posted: 26 Jun 2004, 5:01:04 UTC

The way that people are supposed to use to get around outages and still do useful work is to join other projects. There are currently two that are public with more to follow. Protein folding prediction at http://predictor.scripps.edu/ and the BOINC beta test (ongoing at least until S@H1 is killed at http://setiboinc.ssl.berkeley.edu/ap/.

Heffed
Volunteer tester
Send message
Joined: 19 Mar 02
Posts: 1856
Credit: 40,736
RAC: 0
United States
Message 2286 - Posted: 26 Jun 2004, 6:58:05 UTC - in response to Message 2267.
Last modified: 28 Jun 2004, 21:27:29 UTC

> The way that people are supposed to use to get around outages and still do
> useful work is to join other projects. There are currently two that are
> public with more to follow. Protein folding prediction at
> http://predictor.scripps.edu/ and the BOINC beta test (ongoing at least until
> S@H1 is killed at http://setiboinc.ssl.berkeley.edu/ap/.

I'm trying to attach to predictor and cannot.

I get this:
http://predictor.scripps.edu/ - 2004-06-26 00:38:06 - Sending request to scheduler: http://predictor.scripps.edu/predictor_cgi/cgi
http://predictor.scripps.edu/ - 2004-06-26 00:38:09 - Scheduler RPC to http://predictor.scripps.edu/predictor_cgi/cgi succeeded
http://predictor.scripps.edu/ - 2004-06-26 00:38:09 - Message from server: Project is temporarily shut down for maintenance
http://predictor.scripps.edu/ - 2004-06-26 00:38:18 - Resetting project
http://predictor.scripps.edu/ - 2004-06-26 00:38:18 - Detaching from project

I tried re-attaching to beta, and get the same results. (betas website says project is down for maintenance)

Anybody know why beta being down wouldn't allow me to access predictor either? (Predictor website is up)

I'm tempted to detach from S@H and see if it lets me attach, but I still have two WUs, so I'll wait.

Edit: Nevermind. It works now.

B-SJARMed
Send message
Joined: 13 Dec 99
Posts: 4
Credit: 1,278,002
RAC: 0
Netherlands
Message 2531 - Posted: 29 Jun 2004, 8:15:15 UTC

Darren,

I like your idea of a kind of "reference" packet that has to be send to every computer in the project.
This reference packet can serve two purposes:
First each computer has at least one packet to process.
Secondly from the reference packet is exactly known how many floating point calculations are needed to complete the packet, so the calculation of the credits and the benchmarking of each hostid can be verified and maybe corrected.

For load balancing, it would be beter if the number of packets is low, that requests for more work are checked against number of packets pending for that host. If there is a shortage, queues should be emptied first (so computers with lots of packets queued get no work (or just one WU))
This prevents some computers with 20+ packets queued while other computers starve.

Profile n2vjb
Send message
Joined: 27 Jul 99
Posts: 4
Credit: 35,001,722
RAC: 19,013
United States
Message 10850 - Posted: 21 Jul 2004, 19:03:45 UTC

Now that S@H1 is effectively shutdown (no packet/results/news/???) since 07/16/2004, I have converted all of my windows boxes to BOINC. Now I barely get any WUs to do and this is NO better then S@H1 currently. Take the macine resources from the old project if it is dead and get this one up to snuff. If I ran my systems at work this way and left everyone waiting I would be flipping burgers if I was lucky. So much for using excess CPU capacity to do usefule work!!!

Profile n2vjb
Send message
Joined: 27 Jul 99
Posts: 4
Credit: 35,001,722
RAC: 19,013
United States
Message 10852 - Posted: 21 Jul 2004, 19:05:32 UTC

Now that S@H1 is effectively shutdown (no packet/results/news/???) since 07/16/2004, I have converted all of my windows boxes to BOINC. Now I barely get any WUs to do and this is NO better then S@H1 currently. Take the macine resources from the old project if it is dead and get this one up to snuff. If I ran my systems at work this way and left everyone waiting I would be flipping burgers if I was lucky. So much for using excess CPU capacity to do useful work!!!

John McLeod VII
Volunteer developer
Volunteer tester
Avatar
Send message
Joined: 15 Jul 99
Posts: 24520
Credit: 521,562
RAC: 92
United States
Message 10994 - Posted: 22 Jul 2004, 2:28:24 UTC - in response to Message 10852.

> Now that S@H1 is effectively shutdown (no packet/results/news/???) since
> 07/16/2004, I have converted all of my windows boxes to BOINC. Now I barely
> get any WUs to do and this is NO better then S@H1 currently. Take the macine
> resources from the old project if it is dead and get this one up to snuff. If
> I ran my systems at work this way and left everyone waiting I would be
> flipping burgers if I was lucky. So much for using excess CPU capacity to do
> useful work!!!
>
S@H ran the same WUs out to the crunchers dozens of times to keep everything working correctly. Only 3 are needed for verification. I would hardly call this useful work all of the time. It is still early in the project, and problems are being fixed. However, when there is not data to split, there will be no WUs from S@H. One of the points of BOINC is that it is multi-project, join the projects of your choice. Unfortunately, there is only one other project at the moment: http://predictor.scripps.edu/. I have my queue set to .1 to .2 days, I am a member of several projects (only 2 of which are currently giving out work). I have only had 5 or 6 hours total in the last month where there was no work from anywhere.

Questions and Answers : Wish list : Work unit shortage idea

Copyright © 2014 University of California