"due time"

[7im]

I suspect F@H Burgers are very high in protein

Ouch, that one hurt.

[bruce]

High in protein, but with an appropriate beverage included in the price (depending on whether it's a implicit solvent or an explicit solvent method).

[bruce]

Don't you see that there is a potential to get more processing power for smp-projekts, if you carry longer deadlines?

Yes and no.

The uniprocessor work needs to be processed, too, keeping the appropriate balance across both projects and machines.

If there are more WUs than machines, adding some slower machines does get more work done but at a slower average turn-around time. If there are more fast machines than WUs for the most time-critical projects, then adding some slower machines is detrimental to the project because it increases the project latency.

Remember FAH projects are a combination of parallel assignments and serial assignments. When the serial nature of the assignments dominates the scietific needs of a project (such as the SMP projects), then slower machines are bad for that project. When the parallel nature of the assignments dominates (such as the Uniprocessor projects.

Most undiscovered scientific truths generally lie behind future projects which will be limited by serial considerations, so if there are going to be any changes to the deadlines, they're going to become tighter and tighter, at least in proportion to Moore's Law.

Also remember, the mix of projects and their impairments do change over time. FAH has also added the BigAdv category for highly serialized assignments needing even faster machines than standard SMP, but the scope of those projects is currently lower than it was for a while so those machines are getting mostly standard SMP assignments. Maybe one of these days there might be a lull in the scientific need for SMP projects and from Stanford's perspective, it would be nice if the slower machines auto-magically migrated to uniprocessor projects. (I don't think Stanford's software could do that .... I'm just pondering the possibility.)

Now let's not get into a discussion of "I'm here for the Science" vs. "I'm here for the Points" debate. Both are important, but often to different people.

[stefan42]

okay, looks like it was a good example with the "resturant"

however, it would be nice when the future smp clients could strongly regard the prefered wu-size in the settings. To meet the deadline of 6 days is quite easy with the "small" smp-WUs.

bye,
stefan

[7im]

Smaller WUs get shorter deadlines, so that doesn't help. The deadline is calculated by a formula, just like the Points. The deadline on a small WU would be 2 days instead of 6 days. This information is covered in the FAQs on the project web site. I suggest you read them.

As I said before, this will not change. If anything, the SMP WUs will only get bigger as more and more computers get more and more CPU cores on each chip. For example, at IDF 2010, Intel was demonstrating a 4 CPU system, each with 10 cores AND Hyperthreading. Yes, that's 80 threads! Watch a demo of real time photo realistic rendering...

http://www.youtube.com/watch?v=zbokPe4_-mY