Prioritization of WU Assignments (long)

[Foxery]

I'm feeling curious today, and have a few questions about how WU assignments are handled. I am primarily concerned with uniprocessor CPU clients. Apologies if any of these touch on info you would rather not disclose

Given all the different FAH_Cores, some of which support newer instructions, (SSE1/2) and given that the FAH client is able to detect what CPU I own and what it supports, and given that the client maintains a "Performance Index" based on your actual completion rate, and assuming that bigpackets=yes and -advmethods is set; <inhale>

How intelligently does the server attempt to assign WUs which are "appropriate" to the donor's architecture? Do newer machines strongly prefer newer projects, (especially ones using SSE2 cores,) and older machines (especially ones which do not support SSE2) strongly prefer older projects?

The one hint I've seen about this comes from this line, when fetching new work:
[15:26:10] Initial: 41AB; - Successful: assigned to (171.65.103.162).

With Windows P4s and Core2 machines, the only other value I've seen this show is 40AB. When the client says 40AB, I seem to get higher performance projects. When it says 41AB, I seem to get older projects. One oddity, however, is that if it first says 40AB, and the work server does not respond, it switches to 41AB on the next retry. Is this a bug, or does the client believe that a nonresponsive server means that no "advmethods" WUs are available?



A second, unrelated question:
I think I read on the old forums that you have the ability to tell the assignment servers to prioritize a project when you want a certain one to finish sooner. All of my machines are receiving an inordinate number of p2451/2/3 units lately, though I've also noticed that these have a very large number of Runs/Gens/Clones. Are these prioritized due to running for so long, or are they simply enormous projects?

I've written entirely too much... Thank you for any answers that you are able to give out!

[7im]

Detailed hardware detection was not added until the v6 beta clients. And even now, that detailed information is not yet is use by the Assignment Servers (AS). For instance, if SSE2 work units are the priority, they are still assigned to both SSE and SSE2 capable systems (P3s/Athlons and P4s/A64s).

The project staff currently controls the prioritization of projects to move the science of the project along as quickly as possible.

Within those priorities, the AS routes your client to the Work Servers (WS) by several factors, including but not limited to client type and version, operating system, client switches, minimum performance fraction, and even the number of processor cores for the SMP clients. Then your client will get the next WU in the list on that particular WS. But again, the servers currently do not yet distinguish between a P4 with HT enabled, a dual Xeon, or a Core 2 Duo when counting processor cores.

Note that settings like operating system are not configurable, and Linux vs. Mac clients may be routed very specifically regardless of overall WU priority, as some WU types cannot run on some platforms (yet). The kinks in most fahcores are eventually worked out and released to all types of clients. And note that things like client settings or client switches are only a optional request, not an absolute guarantee of that type of work unit. You may request BigWUs, but if the WSs are out of that type of WU, you will get a normal WU. And even if BigWUs were available, normal WUs might still be assigned if the normals have a higher priority than Bigs.

IMO, cool things with v6 clients are possible in the future, but we still have a ways to go. In the past the number and variety of processors was very limited, so a simple brute force approach to processing was all that was needed. Additional complexity in assigning priorities was not beneficial. But in the future, the assignments will likely get more savy.