Cluster for folding and seti@home!

[pebble guy]

Hey guys i will be building a cluster of 8 optiplex gx620 winth pentium Dś will update with pics and more when i make progress

 

[iHardware Shelden]

Hey guys i will be building a cluster of 8 optiplex gx620 winth pentium Dś will update with pics and more when i make progress

Oh my, are they going to be actually cluster together to form one super computer or run separately?

[pebble guy]

Oh my, are they going to be actually cluster together to form one super computer or run separately?

Yes i will be building a actualk cluster(or thats the plan)

 

Its for a school project

[iHardware Shelden]

Yes i will be building a actualk cluster(or thats the plan)

Its for a school project

Awesome! Ya I have a bunch of those GX620s and I would recommend just leaving the side panels off or flipping the CPU fan to blow out the front of the case They get hot so last thing you want is bad caps because of those wonderful Pentium Ds

[pebble guy]

Sorry been to busy to reply but i will on next friday test diffrent cooling sernarios and give you guys an update. will also try blocking out the space over the heatsink in the cooler shrout, think it might help!!!!

[pebble guy]

also did some reaseach and it looks like i have the small heatsinks without heatpipes. Will look into getting the bigger heatsinks.

[Whaler_99]

You cannot use a "cluster" with Folding at least, and don't think Boinc either. Traditional clusters take the "work", split it up, assign it to free nodes and reassemble when completed. This is usually only for specific hardware. Doesn't work for Folding, mining, etc, at the client level... And you would need to run something like Microsoft compute cluster software, which isn't supported by any clients.

 

Folding and BOINC in of themselves ARE clusters, with the Work servers being the controllers and all our PC's being the individual nodes...

 

Best bet, just install windows 7 on each box, load up the folding software, use the same userID and passkey and you have a traditional folding cluster.

 

Or, install MS's HPCC for fun. It is pretty cool when it works.

[Omnislash89]

Why not Ubuntu?

You cannot use a "cluster" with Folding at least, and don't think Boinc either. Traditional clusters take the "work", split it up, assign it to free nodes and reassemble when completed. This is usually only for specific hardware. Doesn't work for Folding, mining, etc, at the client level... And you would need to run something like Microsoft compute cluster software, which isn't supported by any clients.

 

Folding and BOINC in of themselves ARE clusters, with the Work servers being the controllers and all our PC's being the individual nodes...

 

Best bet, just install windows 7 on each box, load up the folding software, use the same userID and passkey and you have a traditional folding cluster.

 

Or, install MS's HPCC for fun. It is pretty cool when it works.

[brandishwar]

Why not Ubuntu?

Ubuntu out of the box isn't set up for clustering, whereas Microsoft's HPCC is -- it rides on top of Windows Server to distribute computing tasks across a cluster of nodes. You can configure Ubuntu for clustering, but it's not nearly as straightforward.

Going back to @pebble guy, though, @Whaler_99 is right that you can't use a cluster to distribute the workload for any of the distributed computing projects since 1. the clients are supposed to act as the computing nodes of a distributed supercomputer and 2. the clients are not programmed to distribute the work beyond the machine it's running on.

If you're going to build up a cluster of CPUs for Folding or any of the Berkeley projects, the client should be installed on each individual node. Linux will make that relatively easy since the BOINC client packages should be available readily for the major distributions. I have one computer set up for BOINC running at home against three graphics cards. But then those CPUs aren't going to net a lot of points on Berkeley or Folding. It's why the system I'm using runs only GPU tasks.

For keeping those processors cool, though, another option you could consider is water cooling. Build an external water box with a decent pump (probably need something better than a D5) and plenty of radiator space with a manifold and quick disconnects running to each processor through bulkhead fittings. Though trying to find blocks for an LGA775 socket would be interesting.

[pebble guy]

Ubuntu out of the box isn't set up for clustering, whereas Microsoft's HPCC is -- it rides on top of Windows Server to distribute computing tasks across a cluster of nodes. You can configure Ubuntu for clustering, but it's not nearly as straightforward.

Going back to @pebble guy, though, @Whaler_99 is right that you can't use a cluster to distribute the workload for any of the distributed computing projects since 1. the clients are supposed to act as the computing nodes of a distributed supercomputer and 2. the clients are programmed to distribute the work beyond the machine it's running on.

If you're going to build up a cluster of CPUs for Folding or any of the Berkeley projects, the client should be installed on each individual node. Linux will make that relatively easy since the BOINC client packages should be available readily for the major distributions. I have one computer set up for BOINC running at home against three graphics cards. But then those CPUs aren't going to net a lot of points on Berkeley or Folding. It's why the system I'm using runs only GPU tasks.

For keeping those processors cool, though, another option you could consider is water cooling. Build an external water box with a decent pump (probably need something better than a D5) and plenty of radiator space with a manifold and quick disconnects running to each processor through bulkhead fittings. Though trying to find blocks for an LGA775 socket would be interesting.

Cannot watercool then as i dont have the budget, and remember that this i s just  school porject and it is not my computeris but my scholl's

[pebble guy]

You cannot use a "cluster" with Folding at least, and don't think Boinc either. Traditional clusters take the "work", split it up, assign it to free nodes and reassemble when completed. This is usually only for specific hardware. Doesn't work for Folding, mining, etc, at the client level... And you would need to run something like Microsoft compute cluster software, which isn't supported by any clients.

 

Folding and BOINC in of themselves ARE clusters, with the Work servers being the controllers and all our PC's being the individual nodes...

 

Best bet, just install windows 7 on each box, load up the folding software, use the same userID and passkey and you have a traditional folding cluster.

 

Or, install MS's HPCC for fun. It is pretty cool when it works.

http://www.extremetech.com/computing/105951-foldinghome-client-can-now-run-on-supercomputers

[iamdarkyoshi]

I sold a guy like 10 pentium dual core machines for a grand total of 400 bucks, including flat panel monitors and keyboards/mice. All of them with a genuine copy if windon't 7. I think he was going to upgrade his 386 cluster. LOL

[Whaler_99]

http://www.extremetech.com/computing/105951-foldinghome-client-can-now-run-on-supercomputers

 

Did you read all the article? It was for a custom made version of the Folding client, called Copernicus. As far as I know, this was tested and such, but eventually discontinued. The standard FAH client scales out a lot better and is much more powerful, as some have indicated. Would have been way to much money and time to develop it. The other thing, once developed, it would only run on one or two specific computing clusters. Almost all supercomputers have custom built or heavily customized software stacks to optimize their specific work flow.

 

The FAH project has always been about getting to the most accessible hardware. 

[Atomicnerd64]

i say just put them all in a room with the side panels off and enjoy the ppd and heat

[pebble guy]

I red up on cupernicus. it can run on any linux cluster or supercomputer.

THen i was wondering witch os to use for the cluster then. witch distro should i use?????

[bcredeur97]

If you want to setup a cluster for school then I think I would do something else other than folding...

 

I have no real knowledge of clusters and what exactly I would use one for but I have always wanted to get a bunch of RPi's or something and build a cluster.

[BuckGup]

Legit I never thought about this but for my computer class the teacher has about 200 AM2 Athlon X2 computers just sitting there. He also has 2 power edge 2550 server that we can use. Should I hook them up and start folding?

[Imakuni]

Legit I never thought about this but for my computer class the teacher has about 200 AM2 Athlon X2 computers just sitting there. He also has 2 power edge 2550 server that we can use. Should I hook them up and start folding?

If I was you, I'd use Boinc instead. Folding is much more geared towards GPUs, whereas Boinc is more geared towards CPU. Not to meantion, the software allows much finer control, with you being able to completely customize the workload for your specific computers and compute schedule. It also has more variety: perhaps he won't let you help with protein folding, but he may allow you to to math searching (or maybe you'd prefer it that way)?

 

I like to think of it as Windows-Linux to Folding-Boinc; the first is much better known and easier to get started, but once you get a grasp at how the second works, it is superior in every single way.

[BuckGup]

If I was you, I'd use Boinc instead. Folding is much more geared towards GPUs, whereas Boinc is more geared towards CPU. Not to meantion, the software allows much finer control, with you being able to completely customize the workload for your specific computers and compute schedule. It also has more variety: perhaps he won't let you help with protein folding, but he may allow you to to math searching (or maybe you'd prefer it that way)?

 

I like to think of it as Windows-Linux to Folding-Boinc; the first is much better known and easier to get started, but once you get a grasp at how the second works, it is superior in every single way.

I might try bitcoin mining as well just for fun

[Imakuni]

I might try bitcoin mining as well just for fun

Er..... so long as you know you'll be wasting money / electricity, while not helping the society in ANY possible way, be my guest and shia-labeouf.

[brandishwar]

If I was you, I'd use Boinc instead. Folding is much more geared towards GPUs, whereas Boinc is more geared towards CPU. Not to meantion, the software allows much finer control, with you being able to completely customize the workload for your specific computers and compute schedule. It also has more variety: perhaps he won't let you help with protein folding, but he may allow you to to math searching (or maybe you'd prefer it that way)?

 

I like to think of it as Windows-Linux to Folding-Boinc; the first is much better known and easier to get started, but once you get a grasp at how the second works, it is superior in every single way.

 

There are BOINC projects that will use the GPU as well as the CPU -- there just aren't nearly as many of them. The two projects to which I contribute are like that. Folding@Home can also use the CPU. And when first installing the Linux packages and starting the Folding daemon, it will default to using a CPU slot unless you configure it to do otherwise. X2 processors will each pull about 2GFLOPS each -- using BOINC's definition of a GFLOP -- so they'll still do well for BOINC. It'll just be a little sluggish on each task being performed.

[Imakuni]

There are BOINC projects that will use the GPU as well as the CPU -- there just aren't nearly as many of them. The two projects to which I contribute are like that. Folding@Home can also use the CPU. And when first installing the Linux packages and starting the Folding daemon, it will default to using a CPU slot unless you configure it to do otherwise. X2 processors will each pull about 2GFLOPS each -- using BOINC's definition of a GFLOP -- so they'll still do well for BOINC. It'll just be a little sluggish on each task being performed.

I said "more geared", not exclusively. Most Boinc projects are CPU only, and even the ones that do offer GPU tasks usually have more CPU apps than GPUs. But that's not the main reason for me to suggest Boinc over FAH.

 

The main thing is that F@H is much more limited. You have VERY limited control of the software, and pretty much no control at all over the tasks. And the options, as far as research subject goes, are limited to protein folding for a few diseases - and that's it. Since he said he'd be using the school's PCs, maybe he can get his math teacher to support the idea, by contributing to Primgrid. Perhaps, the physics teacher, by helping with particle colision tests (can't remember the exact project right now). Maybe the chemestry teacher is willing to aid, if he helps map molecules for WCG's Clean Energy Project. Who knows, even all of the above, if he splits a few of the PCs for each of them!

 

It's not that F@H is bad, far from that! If that's his final decision, hey, be my guest, thank you for joining the distributed computing community with so much processing power! just feel Boinc is more suited for his particular hardware, that's all.

[pebble guy]

Hey guys sorry that i havnt been able to update you in the last while, i made a desision; im gonna change to ubuntu and use beowulf!!!
gotta be fun an im not experienced in linux!!!

[bcredeur97]

I might try bitcoin mining as well just for fun

you should try a CPU-only coin mining algorithim instead. You'll likely make more.

[BuckGup]

you should try a CPU-only coin mining algorithim instead. You'll likely make more.

Ziftrcoin of EVGA bucks