Prioritizing Projects

[Tobit]

Let me see if I can explain this the way I want it to be received.

Is there a way to know which projects have a higher priority to the scientific community at a given moment and would these be targeted to a specific client running a certain configuration? For instance, if one was folding purely for the benefit of science and didn't care about points, is there one particular client he/she should be running or is running a multitude of clients paramount.

With the new Amber core bonus projects, there seems to be a sense of higher priority for getting these done. Also, Vijay has mentioned that the GPU2 beta has already yielded some impressive findings. However, we all know that there are quite a few folks that push the SMP clients to the limit by running two instances on their Quads in an effort to maximize points above all else.

I guess what I am wondering is, are there certain clients that are being neglected, or under-utilized for sake of a better work, resulting in science taking a back seat to point mongers?

[bruce]

I'm not sure that's a fair question -- or maybe I'm missing your real meaning.

There's enough work for each class of client and they're all doing useful work and can use more CPUs. If you have unused hardware, turn it on and it will be used on one of the important projects that that type of hardware can contribute to. See http://folding.typepad.com/ where Vijay says they're planning to develop the capabilities to run the same code on (almost) any advanced hardware.

The goal is to make the points match the scientific value. That's not always possible, but it is certainly considered as a goal. It stands to reason that work for the "high performance" clients (PS3, GPU2, SMP) is higher priority work than the classic client, and they earn points at a higher rate.

Each of the clients has an efficiency and a range of capabilities. If there is a major shift in either efficiency or capabilities, the points may need to be adjusted, relative to the other clients.

I said it in another thread last evening, but the GPU2 points are still in transition. Initially they were set as a guess, but as the beta progresses, the client is becoming more efficient and points adjustments might be required to keep it on par relative to the PS3 and/or SMP clients. The same would be true if improvements to the efficiency/capability of the PS3 or the SMP clients can be made.

[Foxery]

if one was folding purely for the benefit of science and didn't care about points, is there one particular client he/she should be running or is running a multitude of clients paramount.

With the new Amber core bonus projects, there seems to be a sense of higher priority for getting these done.

My understanding is that the Amber projects are part of some research project that has a shorter deadline than usual, so they want its results ASAP. Normally Stanford is more content to wait for the results, whenever they may come, but all projects were started for a reason, and will be "important"/useful eventually.

After that, Stanford already tells the Work servers to prioritize projects as they see fit.

[VijayPande]

Actually, even within a work server, there is a researcher-defined project weighting (on top of the AS weighting). I usually personally work to balance the servers (which in a sense is balancing resources between researchers in the group) and the researchers themselves balance within their own machines.

We have lots of projects going on at any time, some long term (some of our past calculations took 3 years to run), some short. Some are more experimental (high risk, high payoff), others are more standard (we know it's going to be a useful result, but not as exciting as some of the higher risk stuff). For example, GPU1 was high risk and the payoff is essentially the GPU2 code, which is now going into production. The classic client still is paying critical roles, and we're thinking of new roles for them (such as the special adv AMBER WU's).