Using Queues?

[greenbaarron]

Hello all.

I'm running a GPU folding client on an nvidia 260gtx. Other than the fact that this think is smoking fast I'm having a hard time keeping it fed from the servers. The network I'm on has recently been loosing its comcast connection. I would like the client to download about 6 or 7 work sets so that it can stay running between these down periods (my room got cold last night because the linki was down). Any options I can pass to the client to get it to do this? The network has always come back up within about 5 hours.

[anandhanju]

One of the requirements of FAH is quick turnaround of results, as in most cases the result is used to create the next-in-generation work unit. Having a queue counters this basic requirement and therefore, there is no such feature available.

There has been quite a bit of discussion in this regard and though unlikely, its incorporation into a sebsequent release of the client remains to be seen.

Figuring out why the network connection gets lost may prove to be more fruitful.

[greenbaarron]

Thanks, I was hoping to be more quickly fruitful. The gpu client will just have to waste time then. The problem is going to take weeks to get to the bottom of because of comcast. All the national isps drag their feet pretty bad here in Oakland . I did learn something last night about the GPU/CPU clients and how they interact, and I will be posting that later today to the GPU forum since I think others will find it interesting.

[bruce]

Consider how FAH looks from the server's perspective.

Suppose you have a queue of 6 WUs and you're working on one of them. Nobody can work on any of those other WUs.
Compare that to the no-queue situation. There might be as many as five more computers working on the project so progress will be faster.

Since new WUs cannot be created until the previous WU is completed, there is perpetually a shortage of work to be assigned and when you hoard those WUs, you're inhibiting the scientific process.