An Idea for A Folding Cluster....

[DarkWhite]

I know a guy who has an enourmous surplus (think 100's) of older Dell's, that are decent, like low end core 2 duos and a crappy graphics card if any.
I was thinking of trying to turn these into a cluster for folding.
Ideally these would not be physically changed, and preferably set up so one could be removed and sold if a customer wants one.I have no idea exactly how to do this, but I'm open to suggestions, and I'd really love to see something like that happen, and I'd like to get a little more into folding, as my rig isn't exactly churning out WU's. Plus, those PC's are working for a cause instead of sitting around.
And yes, I know what kind of power those would consume.

 

[Surferseth]

I don't think there would be a way to cluster them, but you could run the SMP client on each dual core system.

 

[ChasR]

I think you can give up on that idea. SMP2 is close. It's going to be thread based as opposed to process based that clustering needs. You'd need a 100 Gb network to make it work with the current SMP client if there were a way. It would be a lot of work for a little gain. With the design of the FAH client, clustering is unnecessary anyway. All those old dual cores will do as much or more work individually as they would as part of a cluster, with no cluster OH. Connect them to a network and install a minimal linux distro like notfreds on each and let them rip. A 2.0 GHz C2D makes about 1500 ppd on current a2 SMP WUs. You can do the math and decide if it's worth doing.

 

[dark bishop]

I'd pick up a bunch of cheap 1Gb usb flash drives and install the NotFred usb client on them.

 

[ozzlo]

Make sure you remember that if the guy runs 100's of computers, his power bill is going to skyrocket...

 

[AmbientFiction]

Make sure you remember that if the guy runs 100's of computers, his power bill is going to skyrocket...

But Ozzlo this is a great tax write off. lmao. Count your power bill as a donation.
Hopefully it will slide by.
< Before they catch ya.

 

[DarkWhite]

Okay, it was just an idea, I was just thinking distributed client=better with more computers, but if that's not gonna work then I'll talk to him about just running individual versions.
Thanks for the help, even if it was the opposite of what I was hoping for

 

[dark bishop]

Okay, it was just an idea, I was just thinking distributed client=better with more computers, but if that's not gonna work then I'll talk to him about just running individual versions.
Thanks for the help, even if it was the opposite of what I was hoping for

Your thinking wasnt to far off except you were looking to small, the entire folding project is a giant cluster in itself, any computer we run a folding client on is one node of the cluster.

 

[Shiggity]

Start off running a few machines and go from there

But building a cluster out of all that hardware would be a fun project. There are a lot of scientists out there with data sets looking for horsepower.

 

[DarkWhite]

I'll probably talk to him about getting a few machines running, if DB is right then I see why a cluster would be useless.
It'll be kind of tedious getting all those machines running seperately, especially as they don't have their own monitors.

 

[dark bishop]

NotFred has a lot of really good options to use on headless machines that are simple enough to set up and are fairly high performance.

 

[Adak]

That is how a Stanford FAH programmer (Kassom, iirc), expressed it, in explaining why they don't offer a cluster client.

You might want to take this up with the Stanford FAH group, in the FAH forum. It has been brought up (and shot down), several times before, but you know what they say about squeaky wheels getting the grease.