Folding Forum

Please, sir, may I have another.

J - still pulling the 45xxs strong.

AgrFan and others:

As you may have noticed, the -advmethods AMBER WUs have been on low priority in August, as our third-party prediction collaboration has been drawing to a close. However, before we finally close this chapter, we are planning in the next week or so to run a series of benchmark simulations with known sequences. The calculations will be a bit less urgent, because we will not have deadlines this time. The assignment weight will be lower accordingly.

Expect an update on this soon. More info to come in the next week --Vince

Vince,

Thanks for the update!
I figured the bonus Ambers were starting to run out when I started getting 5113 Gromacs33 WUs last week.

Please let us know when the new WUs are available ...

AgrFan

Hello -advmethods folders:

There is a "last hurrah" of bonused -advmethods AMBER WUs coming online: projects 4597 through 4623. These will start appearing in the next hour, trickling fully online throughout the day. They will continue to run over the next month or so. These projects are fragment simulations of known sub-millisecond folders, the results of which we can compare against the simulations earlier this summer.

As before, see http://fah-web.stanford.edu/cgi-bin/fahproject?p=4597 for points.

Thanks again to all the -advmethods folders helping with this effort.
--Vince

Out of curiousity, why are many of the Amber units extremely small? 47 pointers take a few hours to crunch, even on fairly old systems, yet have deadlines many weeks long.

Because this Project is time sensitive they want to get as many results back as possible I bet.

Baowoulf wrote:Because this Project is time sensitive they want to get as many results back as possible I bet.

... WAS ...

but the time-sensitive portion is ending, as Vince said.

but the time-sensitive portion is ending, as Vince said.

To clarify: These latest simulations are non time-sensitive per se, but we need to run them with the same protocol as the projects this summer (same WU lengths and deadlines). The only difference will be the assignment weights.

Vince

I stand corrected my bad.

Vince,

How can we get the new units consistently like we were this summer? I pulled another 5113 unit tonight due to the higher weighting for 5113 than the new 45xx-46xx units. 5113 takes 3.5 days on my machine which is way too long for me ...

The minimum client version for 5113 is v6.20. The minimum client version for the new 45xx-46xx units is v5.04. If I run the v5.04 client will that help?

Agrfan

BTW, how do you read the comma separated values in each column on the server status page? The values appear to correlate to the projects being serviced from each server. The number of values doesn't seem to match the projects listed on the project summary page.

Er, I'll spell it out.

1. Why are the WUs so small?
A) This generates a lot of network traffic for Stanford.
B) This causes small amounts of downtime for large numbers of CPUs. It adds up.
C) This requires the Work Servers to spend a lot of time processing results and generating new WUs.
D) Maxizing resources requires keeping an enormous number of "operations in flight" at all times, adding unnecessary complexity to administration.
E) If there are 80 projects being processed simultaneously, the researcher sits around for several weeks with no results, and is then suddenly inundated with all of them at once. This doesn't seem favorable for the humans reading it all and writing up a report.
2. Why are the deadlines so long?
A) Results are said to be time sesitive. The slowest of CPUs finish these in a day. Yet the deadlines are 2-4 weeks.
B) Any time a WU is lost, it therefore takes 2-4 weeks to be reissued. The time sensitivity is harmed.
C) See also Part "D" above.

I'm happily crunching these on several machines. Just wondering if the efficiency could be improved.

AgrFan wrote:Vince,

How can we get the new units consistently like we were this summer? I pulled another 5113 unit tonight due to the higher weighting for 5113 than the new 45xx-46xx units. 5113 takes 3.5 days on my machine which is way too long for me ...

When the bonus was introduced for the Amber core WUs, we had a clear warning that they would be temporary. If you've gotten used to that temporary condition, it's time to scale back your expectations, because a situation like that probably won't come around again for a few years. If you're worried about the points race, don't be concerned. The same thing is happening to everybody.

A 2.8 GHz P4 should be getting 110 points per day, just like the FAQ says.

1. Why are the WUs so small?

All your points are correct, but consider the task this summer -- we had to turn around lots of WUs in about a week or two, and insure good coverage for many different fragments. Yes, the 8-mer fragments finish quickly, but the 16-mer fragments can take up to day on some machines. Because even days were precious, we opted for shorter WUs

2. Why are the deadlines so long?

Mostly, we were following the standard formula here, because we wanted to give everyone credit with the bonus points. Also, lost WUs were generally not a problem, as there usually were so many unsent clones.

I'm happily crunching these on several machines. Just wondering if the efficiency could be improved.

I agree it could be improved, but we really need to run these with the same protocol, to get a similar distribution of samples.

Vince

Cool, thanks for the insight

codysluder wrote:

AgrFan wrote:Vince,

How can we get the new units consistently like we were this summer? I pulled another 5113 unit tonight due to the higher weighting for 5113 than the new 45xx-46xx units. 5113 takes 3.5 days on my machine which is way too long for me ...

When the bonus was introduced for the Amber core WUs, we had a clear warning that they would be temporary. If you've gotten used to that temporary condition, it's time to scale back your expectations, because a situation like that probably won't come around again for a few years. If you're worried about the points race, don't be concerned. The same thing is happening to everybody.

A 2.8 GHz P4 should be getting 110 points per day, just like the FAQ says.

I get the same PPD with 5113 as I do with the 45xx units. I like to send work back as quickly as possible. If I can send a unit back in 18 hours or less, I'd prefer that route. Too much can happen in 3.5 days to cause a unit to be lost.

I just pulled a 46xx unit running the v5.04 client ... Project 4619, 174 PPD