Building a farm of CPUs

Note: this may have been thought of, done, or have a thread here, but I haven't been able to find a like to my question.

Relatively new to folding, but I have an extra rig here (not running, inefficient)and the possibility of getting a lot of cheap dual cores and mobos in the future.

Is there an economical way to run these for folding by linking them to a common OS? I'm thinking about making a tower for this type of setup, but running all of these seperately could be burdensome. Space and cost are primary concerns, I would like to minimize these while folding (otherwise the cost isn't an issue).

This is rough draft of a thought, so there are probably things I am overlooking. Wondering about the feasibility of the idea.

Just the thought of all these poor old dual cores with so much left to give...

jmsanders2 said:

Is there an economical way to run these for folding by linking them to a common OS? I'm thinking about making a tower for this type of setup, but running all of these seperately could be burdensome. Space and cost are primary concerns, I would like to minimize these while folding (otherwise the cost isn't an issue).

Click to expand...

Someone will need to provide links, because it's too late right now for me to properly remember. But people do something like this, yes. Each um, thingy, only has a motherboard, cpu, ram, power supply and network connection. No graphics card, sound, harddrive, keyboard, mouse, monitor etc. Each of these thingies connects to some fully built, central box and it somehow manages all these barebones setups for folding, by way of the network. It's pretty cool, I want to try it myself someday.

There's some kind of detailed setup guide to do this either here, or linked from here.

Its called network booting... it can be pretty handy.

Each PC that boots over the network via files stored on a server are called nodes...

Dig around the stickies for a while looking for them. I'm suffering Sorin's problem at the moment... its 5:30 and I'm pulling yet another all nighter getting ready for finals next week.

Well I'm glad you asked because I'm in the middle of big experiment with clustering CPU's up.

A) Hows your linux skills?
B) Do you have a gigabyte network? (I'm only using a 100mb and its a killer)
C) Up for a challenge?


Most people just network boot their machines, meaning that every machine looks to the center machine and loads its OS off of there and folds that way. What I propose is a cluster where all the machines combined look like 1 in 1 OS. Imagine with 5 dual cores your OS (Linux) see's it as a 10 core machine! Just think of the folding possibilities!

I believe what you are talking about is referred to as a Beowulf Cluster. It is a method to network multiple computers with shared resources to tackle a complex problem. It is like a localized distributed processing method to build affordable "supercomputers".

icantfindone said:

I believe what you are talking about is referred to as a Beowulf Cluster. It is a method to network multiple computers with shared resources to tackle a complex problem. It is like a localized distributed processing method to build affordable "supercomputers".

Click to expand...

That is precisely what I am talking about. The thing is though unless you have native 64bit CPU's you can't use the Linux SMP client. So I'm working on getting either QEMU + KVM or Xen (with ReactOS to run Windows SMP) and Kerrighed (clustering linux kernel) working in sync. Then you can have up to a 255 core cluster adn run multiple windows SMP clients.

Shelnutt2 said:

Well I'm glad you asked because I'm in the middle of big experiment with clustering CPU's up.

A) Hows your linux skills?
B) Do you have a gigabyte network? (I'm only using a 100mb and its a killer)
C) Up for a challenge?


Most people just network boot their machines, meaning that every machine looks to the center machine and loads its OS off of there and folds that way. What I propose is a cluster where all the machines combined look like 1 in 1 OS. Imagine with 5 dual cores your OS (Linux) see's it as a 10 core machine! Just think of the folding possibilities!

Click to expand...

Thanks to all for the replies and info guys.

I'll be looking into this further as a project now with this info as a start. One thing that caught my eye is B) Do you have a gigabyte network?

Is this necessary with smp folding many cpus? My initial thought was the computer only contacted stanford for a WU and then transferred the data back when completed...if this is not true and a large bandwidth is needed then this would be good news to know.

Another thought came to mind, are there tax breaks at all for large folding as 'charity'? This would be amazing if Stanford could arrange something something like this as it is measurable (although verification and security might hinder this). On hind sight, this would require a lot of work on Stanford's part. My morals have prevented me from ever giving blindly, but folding is a way to contribute with a clear conscious.

I can dig deeper and research the rest as things come fruition. This will be an ongoing pet project as of right now.

shel i follow you and i asked a while back why can't we use clustering and have each cpu looked at as one? i mean why not have linux break down the WU, send out each section to a cpu. the more cpus you have the smaller the section of the WU is sent to each cpu. imo that would vastly increase folding time then running 10 seperate ones or am i just thinking out of my

http://www.ocforums.com/showthread.php?t=436219&highlight=windows+diskless+folding

windows diskless folding thread
works for linux too