• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Building a Folding Farm

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.
Actually my Farm is in aurora, shoot me a pm if you want to take a look, we can take a pic of you next to it as proof of life :rofl:
haha well thats even closer, Im over in littleton/centennial. Certainly would like to come see it sometime but Im moving into my house this weekend. Ill hit ya up on pm this weekend. :rofl:

Always like to see other people's setups, especially in person :thup:
 
Ever think about making a cluster out of those? The SMP client maxes out at 128 threads. Just dedicate 16 systems to the cluster. Use Linux-HA as the base http://en.wikipedia.org/wiki/Linux-HA then run Windows server 2008 in a VM. I believe the wu bonus grows exponentially so 128 threads crunching away on the same wu should give you a much bigger bonus than the 16 separate systems working on their own bigadv wu's.

:thup::comp:
 
Ever think about making a cluster out of those? The SMP client maxes out at 128 threads. Just dedicate 16 systems to the cluster. Use Linux-HA as the base http://en.wikipedia.org/wiki/Linux-HA then run Windows server 2008 in a VM. I believe the wu bonus grows exponentially so 128 threads crunching away on the same wu should give you a much bigger bonus than the 16 separate systems working on their own bigadv wu's.

:thup::comp:

I'm pretty sure FAH can't run on a cluster. I remember a lot of discussion about this ~1 year ago, and basically the cores have way too much intercommunication for a cluster to work well, because the individual computers in the cluster would have to talk so fast. Correct me if I'm wrong, but I'm pretty sure that's the current state of things...

Also :comp::comp::comp: Crazy farm sfu
 
Ever think about making a cluster out of those? The SMP client maxes out at 128 threads. Just dedicate 16 systems to the cluster. Use Linux-HA as the base http://en.wikipedia.org/wiki/Linux-HA then run Windows server 2008 in a VM. I believe the wu bonus grows exponentially so 128 threads crunching away on the same wu should give you a much bigger bonus than the 16 separate systems working on their own bigadv wu's.

:thup::comp:

Has anyone actually made it work, if it would I would be game, but I am not sure if it works. Hmmm I am going to have to do some research.
 
I'm fairly sure that clustering wouldn't work as you want. Cluster programs have to specifically written to work in cluster environments. The data transfer to each folding process will exceed any home network, by far. If there was a cluster version, it would send the work to the node to process and it would simple return the results. Otherwise, you are going to be transferring memory-space across networks; and networks are substantially slower than the memory bus (7-10gb/sec compared to 125mb/sec).

-----

On the other hand, I'm thinking about working on a PXE/other environment with saving to network devices (running linux as the client OS, of course). I'm doing this since I can't fold due to money constraints. This PXE/other environment would allow your nodes to run the OS without a hard drive. It simply pulls an image across the network and starts folding. This would lower energy costs by a small amount and lower setup times by a substantial amount. Adding another client would be as simple as plugging it into the network and hitting the power button. I've just started the "thinking about it" stage, haven't tested yet.

This thread that you started got me thinking about it.
 
I'm fairly sure that clustering wouldn't work as you want. Cluster programs have to specifically written to work in cluster environments. The data transfer to each folding process will exceed any home network, by far. If there was a cluster version, it would send the work to the node to process and it would simple return the results. Otherwise, you are going to be transferring memory-space across networks; and networks are substantially slower than the memory bus (7-10gb/sec compared to 125mb/sec).

-----

On the other hand, I'm thinking about working on a PXE/other environment with saving to network devices (running linux as the client OS, of course). I'm doing this since I can't fold due to money constraints. This PXE/other environment would allow your nodes to run the OS without a hard drive. It simply pulls an image across the network and starts folding. This would lower energy costs by a small amount and lower setup times by a substantial amount. Adding another client would be as simple as plugging it into the network and hitting the power button. I've just started the "thinking about it" stage, haven't tested yet.

This thread that you started got me thinking about it.

I had a PXE boot mini farm running on overclocked Thunderbirds back in 2003. You can do it. Only one machine had a hard drive. The other four had 128mb of memory and cpus only. Can't remember the board but they didn't even have a built in video card. I even used a 20 pin ATX splitter to share power between two boards. Wish I could find some pictures of that thing, it was pretty fun. Kept the basement nice and warm. :)

When I retire and get my ground source heat pump and dedicated solar cells for power, I will bring back a new version of the farm.
 
I know it can be done. The main problem is saving the progress. On the single core clients, that doesn't matter much. On the SMP clients, it is a huge point loss if a system goes down (or the entire farm loses power). If someone didn't care about the risks, it would be simple to setup a PXE server and boot a linux image with everything setup and ready to go. I was thinking more along the lines of saving the progress, but that adds substantial complexity. How would you tell the difference between each unit since there is no "setup" and how would progress be saved?

--------
EDIT (revelation):
I actually just figured a way to do it as I was typing this. If you were to have a share on the PXE server (Samba, etc), the client could use its own MAC address as the folder name. This would allow a unit to pick back up where it left off.

I was originally thinking of a way to see if a folder/file was "locked", but that is a tad more difficult and requires a lot more work to find a "not in use" folder. Using the MAC address would be extremely simple.

I'll mull this over a bit and see if I can get a prototype working. It might be a while since my next goal is running a cluster. The OS's are installing now; just got all the power and data cables ran.
 
Last edited:
Thanks for the links. I'll see if I can take those into consideration.
 
Err, and I guess I'm just a bit slow. NotFred's disk does this by IP address if you are running a TFTP server. This simplifies my research by a ton.
 
I am familiar with the diskless folding but it is linux only, so unless someone knows how to make notfred's rune wine, I am not seeing the upside. Could be I am missing something here?
 
Back