Time to drop the 2nd client?

[Kendrak]

It seems on both of my Q6600's that I'm getting WU in the last few days to need the whole CPU to finish by the preferred deadline.

I currently have a massive 3065 worth 2144 pts. It seems to have slowed the other instance down so that it will now not finish by the preferred deadline.

I'm still finishing with room to spare on the final deadline.... but I head somewhere that Stanford likes to get the WU back in time.

What are your thoughts.

 

[aldamon]

It's a big ethical problem for sure. I don't believe there is a penalty for missing the preferred deadline, but how much does it affect them if you miss it be a couple of hours? Why is the difference between the final deadline and preferred deadline so large on these new units?

Clearly Stanford still prefers brute force over efficiency with Kentsfield if they're going to insist on 24-hour preferred deadlines. I think that's nuts, personally. If you need to use the computer, a 24-hour preferred deadline gives you very little cushion. If we complain, they'll just insist they want units back as fast as possible. We all know they're throwing away efficiency that way. Oh well. If this is the direction they're going, I won't miss running VMWare. It's a PITA sometimes.

Honestly, I would keep the dual-clients until these kind of work units are the only ones left. I got them for a couple of days and then the 2605/2653s starting flowing again. My machines at home are working on four 2653s of them as I type so they're still out there.

EDIT: Just got another 3064 on one client. Prefered deadline: 24 hours / Final deadline: 48 hours. 1329 points again. NUTS.

 

[Killer[MoB]]

If the machine will not make the preferred deadling on the work units, then you should drop the second instance. Once a WU passes the preferred deadline, the WU is issued to someone else to make sure it is getting done. Basically causing the same WU to be folded twice. It's a waste of time science wise.

 

[aldamon]

That stinks. What's next, 12-hour preferred deadlines? They're being kind of greedy here.

 

[Bill Clo]

Part of the reason I am giving up GPU folding. Too much power invested in too little return. The freaking client is stll in Beta a year and a half later. I now save $25 a month in power which can now be diverted into my other main hobby. Plus whatever I can get for selling those power hog GPU folders.

I also think Stanford is forgetting who pays the bills here, and have forgotten where their bread is buttered.

 

[aldamon]

...............

 

[aldamon]

Complaints:

http://foldingforum.org/viewtopic.php?f=8&t=1604&p=14415#p14415

Possible explanation for now:

by uncle_fungus on Mon Mar 17, 2008 8:20 pm
Also, check the deadlines reported by FahMon. I've been alerted that the current version has problems with non-integer deadlines which will affect the 306* series of WUs since the real deadline is 2.2 days, but FahMon incorrectly reports it as 2 days.

Maybe we're not actually missing the preferred deadline?

EDIT: Confirmed!

3064 171.64.65.63 p3064_lambda5_2003 21521 Preferred Deadline: 1.80

http://fah-web.stanford.edu/psummary.html

 

[Kendrak]

The 2nd instance LIVES!!!!!

 

[Mohonri]

On the topic of efficiency, I think we need to look beyond the scope of a single computer. The reason Stanford wants those WUs back fast is because certain WUs "unlock" further WUs. I get the feeling that currently, they have a surplus of raw computing power, and they can't keep up. In other words, clients are requesting WUs faster than Stanford can generate them. I guess they must be sending out WUs to multiple computers to keep them all busy. A single SMP WU, for example, may unlock dozens or hundreds of other WUs. If you run a single client on your quad core and get the results back, say, 8 hours earlier, than Stanford can create and send out those dozens or hundreds of new WUs to computers that would otherwise be duplicating work. So you take a (relatively) small PPD hit, but the benefit to the project as a whole is much greater.

This is, of course, pure speculation, but it makes sense. Now, if we could somehow arrange things so that unlocked WUs would go to the same team that turned in the WU that did the unlocking, that would be nice, but that's getting out in the weeds now...

 

[Pocatello]

could someone post a cliff notes version of what is going on? Or something that might make sense to a dummy like me? thanks.

 

[Kendrak]

FAHMon makes the deadline on some WU look like they are only one day. They are in reality 1.8 days.

Every thing is good in the world.

 

[BillR]

3064 171.64.65.63 p3064_lambda5_2003 21521 1.80 3.60 1753.00 100 GRO-SMP Description densign

Folding at home monitor is an interesting program. Its PPD count is pretty much bogus and, by relying on second and third party info some of you are misinterpreting a lot of what the program is saying.

The preferred deadline is when Stanford would “Like” to have the work back, it is not the final day they will accept it back.

On a 3064 you have 1.80 as the preferred but 3.60 days before the deadline.

I’m not at all sure how with a Q6600 with a mild OC you can’t hit under 24 hours let alone 1.8 days and that’s running dual SMP clients on the Quad.

Please check the source for the real information:

http://fah-web.stanford.edu/psummary.html

A monitoring program is a great way to be sure everything is running, no doubt about it. What we are all really interested in is our actual production. Of course that can be found at www.hardfolding.com. If you look at your daily average in PPD or PPW those are the real numbers, in fact every combination of every calculation can be found only a mouse click away, ok, a few mouse clicks away.

I too hate high electric bills however my current output is the result of only 4 machines.

Don’t worry, be happy and above all keep on folding