Same work unit sent to how many clients?

Message boards : Projects : Same work unit sent to how many clients?
Message board moderation

To post messages, you must log in.

AuthorMessage
darwincollins

Send message
Joined: 9 Jan 10
Posts: 18
Message 30664 - Posted: 25 Jan 2010, 5:23:31 UTC
Last modified: 25 Jan 2010, 5:25:14 UTC

I have read messages about the same work unit being sent to mulitple clients, and using a quorum of 3 in order to validate the result.
Here is one of the threads that I read:
http://aqua.dwavesys.com/forum_thread.php?id=72

The message is several years old, but, i imagine even in these modern times, you probably have x initial replications and need a x for a validated results.

Is it the same for each project? How can we find out how much of our work is duplicated on other accounts just to come to the same conclusion?

Or is boinc smarter now so that we don't waste time working the same work unit that x other computers had already completed and was accepted.
ID: 30664 · Report as offensive
mo.v
Avatar

Send message
Joined: 13 Aug 06
Posts: 778
United Kingdom
Message 30676 - Posted: 25 Jan 2010, 13:54:43 UTC

Hi Darwin

It's different for each project; each chooses its own settings for how many tasks must be completed from each workunit.

You can see the situation for the tasks you're processing by going to your account for the project in question, then going through the link to your computer and then to its tasks. Each task belongs to a workunit and you can click on that link. You'll see what the settings are and can look at who else is processing the tasks in the workunit.

I believe there are Boinc plans to allow projects to identify particularly reliable computers whose results are almost always successful and almost always validated. The projects would if they want be able to send to these computers tasks which wouldn't be duplicated. Occasionally these computers would process a duplicated task as a check. I don't know how far this plan has been developed but it seems to me a good idea in that it would reduce the amount of task duplication.

Inevitably, though, as projects use unknown computers belonging to unknown people they must check whether the results are good before researchers use the data.
ID: 30676 · Report as offensive
SekeRob

Send message
Joined: 25 Aug 06
Posts: 1596
Message 30679 - Posted: 25 Jan 2010, 14:41:28 UTC - in response to Message 30676.  

I believe there are Boinc plans to allow projects to identify particularly reliable computers whose results are almost always successful and almost always validated. The projects would if they want be able to send to these computers tasks which wouldn't be duplicated. Occasionally these computers would process a duplicated task as a check. I don't know how far this plan has been developed but it seems to me a good idea in that it would reduce the amount of task duplication.

Inevitably, though, as projects use unknown computers belonging to unknown people they must check whether the results are good before researchers use the data.

World Community Grid [WCG] has been doing this for years through customizations, but still random sampling quorum 2 is done on a small percent to maintain that scientific very high rate of confidence. That can of course only be done when the result has no absolute value such as in simulations, when thousands of results are used to find common convergence points. In cases like Help Cure Cancer and Help Cure Muscular Dystrophy the energies and analysis go into a reference database. Every bit must be correct and thus verified. Like the old days double entry was used to make sure that no GO developed from the GI.

To get into the reliability group is not difficult. 15 results consecutively verified and validated for a science... one false move and it taking another 15 or so to get back in. Increases project efficiency dramatically.

Coelum Non Animum Mutant, Qui Trans Mare Currunt
ID: 30679 · Report as offensive
darwincollins

Send message
Joined: 9 Jan 10
Posts: 18
Message 30700 - Posted: 27 Jan 2010, 4:39:03 UTC - in response to Message 30679.  
Last modified: 27 Jan 2010, 5:07:32 UTC

I like double entry. I like ECC memory too.

However, an excess duplication of work also means that the code is as 'unknown' as the computers that they run on. right?

Hopefully, there are some really smart math/process guys optimizing these schedules, and its not micromanaged by researchers. grin

I did not have much luck on following the links to see what the settings the various projects use. Well, ok, found the roseta stats. Dang... lots of stats.

Oh I did see a compute error had occurred, but, also saw that another computer had the same client error on the wu. cool
ID: 30700 · Report as offensive
SekeRob

Send message
Joined: 25 Aug 06
Posts: 1596
Message 30704 - Posted: 27 Jan 2010, 8:57:07 UTC - in response to Message 30700.  

FightAIDS@home has done over 100 million results. Originally it started out as quorum 3, then became quorum 2 and now for about 2 years been running zero redundancy, single copy distribution. Similarly, from the outset Discovering Dengue Drugs, Influenza Antiviral Drug Search and the current Help Cure Childhood Cancer run in this modus... and much faster than the researcher ever hoped for. ETOC mid 2011, originally thought to last well into 2012... so maybe they'll add more targets.
Coelum Non Animum Mutant, Qui Trans Mare Currunt
ID: 30704 · Report as offensive

Message boards : Projects : Same work unit sent to how many clients?

Copyright © 2022 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.