Deadlines seem strange and illogical.

This forum is for discussions about the Motherboards.org Folding team. What is folding? Venture on in for a look.

Moderator: The Mod Squad

Deadlines seem strange and illogical.

Postby Daft Ada » Thu Jan 17, 2008 11:59 am

When I was running SMP, I struggled with the deadlines being 3-4 days depending on the exact work-unit. Running non-SMP, it's the complete reverse. I'll be submitting a work-unit in about 6 hours time, but even the preferred deadline is not for another 38 days! :o

If they can be that relaxed on the non-SMP, why can't they relax the SMP a bit? :?
Daft Ada
Black Belt 1st Degree
Black Belt 1st Degree
 
Posts: 1358
Joined: Sun Nov 12, 2006 11:03 am

Postby Pette Broad » Thu Jan 17, 2008 4:24 pm

Those long deadlines were set in the days when a 500 Celeron was considered to be a good machine :) . I mean you could download a "standard" unit, switch the machine off, go on a world cruise and still get them in on time. I don't see how they'll ever get some of the projects finished, for example if someone downloads one of those giant gromacs33 units, doesn't fancy it and deletes it, then it won't be reallocated for several months and if the next person deletes it.............

I don't understand why units have to be so gigantic, they should be split up into smaller units like Rosetta@Home does. I've looked at all the Distributed projects, and there's dozens of them, Folding@Home is only second to Climate Prediction in the time it takes to do WU's.

Don't know why SMP deadlines aren't raised a tad, 24 hours extra would make a big difference....but maybe they don't want more users :?

Pete
Image
Pette Broad
Black Belt 5th Degree
Black Belt 5th Degree
 
Posts: 5490
Joined: Tue Jul 10, 2001 12:01 am
Location: Flintshire, U.K

Postby Karlsweldt » Fri Jan 18, 2008 7:22 am

May also be a cognitive recognition of an accelerated folding process.. which has not gone beyond the 'experimental' stage with its results. Consecutive identical results are critical to the determination of the protein folding process.
If you get an occasional WU of this type, perhaps allowing a 24/7 run until completion would be best. For all others, use a nominal 16 hours active with 8 hours off. I have no problems with making any deadlines by about 50% of minimal time limits that way. Still haven't had one of those "over-nighters".. but if it comes, I will run my setup for its duration.
If you get a frequent offering of that type, then reconfigure your Console to not accept that variety. But if they have a large point value, then.. well, your choice!
F@H.. to solve mankind's maladies.. in our lifetimes!
Karlsweldt
Mobo-fu Master
Mobo-fu Master
 
Posts: 20663
Joined: Wed Nov 12, 2003 11:57 am
Location: 07438

Postby Pette Broad » Fri Jan 18, 2008 6:58 pm

Karlsweldt wrote: Consecutive identical results are critical to the determination of the protein folding process.


That's another oddity about Folding@home, duplicate WU's aren't sent out, they just rely on the one unit being right. In the past I've dabbled with a few Boinc Projects, like Rosetta@Home, World Community Grid and they always send out at least 3 identical units (some projects send out 4). If the results from one of those units differs then another one is sent and so on until they get the required number of matches. Maybe Folding@home has a better validation system 8)

Pete
Image
Pette Broad
Black Belt 5th Degree
Black Belt 5th Degree
 
Posts: 5490
Joined: Tue Jul 10, 2001 12:01 am
Location: Flintshire, U.K

Postby MrCraigCraig » Sat Jan 19, 2008 8:28 am

I think F@H does distribute multiple redundant projects.


http://folding.stanford.edu/English/FAQ-main#ntoc11
Running


I just finished a WU and now I got another for the same protein. Is there something wrong?

No, everything is fine. We're studying the dynamics of several proteins, so you may receive work units from the same protein (and same project number) multiple times. Each WU gives us additional information about the dynamics of that protein, so it is important to us. Indeed, if we did only 1 WU per protein, we would not learn very much. A Project number, and a Run, Clone, and Generation number distinguish each work unit.
I tried using a "Point of View Gun", but my aim was lousy. ;-)

ASUS K8N, AMD Sempron 3100+(1.8ghz), 1gb RAM, eVGA 5500FX 256mb AGP, two WD120gb SATA drives in RAID 0 (striping), 450w power supply.

Fold well, so that others may prosper.
MrCraigCraig
Black Belt
Black Belt
 
Posts: 887
Joined: Fri Jun 03, 2005 9:33 pm
Location: NH

Postby Pette Broad » Sat Jan 19, 2008 9:25 am

MrCraigCraig wrote:I think F@H does distribute multiple redundant projects.


. A Project number, and a Run, Clone, and Generation number distinguish each work unit.


That's the important wording above. Each project has a Run, Clone and Generation number, so for example Project 891 Run 10, Clone, 1, Gen 1 will only be sent to one machine. When this unit is returned completed to the server it'll generate the next unit (Gen 2). A duplicate is usually only sent when the preferred deadline for the first machine is passed or when a partial WU is returned. On rare occasions a genuine duplicate will be sent out, this happens when 2 machines access the server at the same time.

Also, the server flags each unit and records the Machine ID, so if you have an instant Early Unit End with a client/core error the server will send you the same WU again, this will also happen if you delete or "lose" a unit.

This is a better explanation.

If you DO have the exact same WU on more than one machine, something is seriously broken. The servers know which WUs they've given out. They then wait for each individual WU to come back in. If they do not come back in enough time, that WU is sent out again and you are denied credit for it. If it turns out that you've got 3 identical WUs the servers are broken and you're not going to get credit for more than one. That is designed so that you can't get a good WU from a project like p1481 or p1141 then copy it over to all your machines.



Pete
Image
Pette Broad
Black Belt 5th Degree
Black Belt 5th Degree
 
Posts: 5490
Joined: Tue Jul 10, 2001 12:01 am
Location: Flintshire, U.K

Postby Karlsweldt » Sat Jan 19, 2008 10:27 am

Each WU has a 3-point ID, same as each client (user). This avoids duplication of WUs for extra credit that is not intended. The servers have records of what WUs were issued to whom, and when.. and the FAHlog.txt has notations about each WU process in effect. The client has a unique machine ID # and user ID plus team ID.

The F@H process is serious business.. intended for research into genetics processes and curing mankind's maladies. Counterfeit or cloning of processes can lead to false results!
F@H.. to solve mankind's maladies.. in our lifetimes!
Karlsweldt
Mobo-fu Master
Mobo-fu Master
 
Posts: 20663
Joined: Wed Nov 12, 2003 11:57 am
Location: 07438


Return to Motherboards.org Folding Team

Who is online

Users browsing this forum: No registered users and 1 guest