Distributed Computing Thoughts/Plans

[TC10284]

Hey all,

I wanted to lay out some thoughts/theories/plans here since this forum sounded appropriate. I would like some opinions. I have a lot of thoughts, but bear with me, please...

The Details:


* I have some old spare systems; probably 150 to 200. Most are homogeneous Compaq Pentium III 933MHz systems. I also have a lot of homogeneous IBM NetVista P3 800 and 933MHz systems. They are still at my parents house doing nothing. They are complete systems; onboard Intel video, 128 to 256MB PC100/133, 10 - 40GB HDs, with CD-ROM drive. I think these would suit well for setting up each as a BOINC, or F@H client under Fedora. They may not be the fastest things, but they would compute for science/health/astronomy and that's all I'm concerned about.
I have also tinkered with the idea of setting up a Linux cluster using OSCAR or similar.
Setting up the computers won't be a problem at all.

* I would like to keep each BOINC "cluster" (set of machines) and Linux cluster behind a VLAN or router to keep any broadcast traffic to a minimum. Making another router out of one of the non-homogeneous systems is not a problem (IPCop, BrazilFW, etc.)

* I also have spools of unterminated fiber that I can use to run underground to the building to the basement for uplink to my primary network in the house. Not to every system, but more of a "long-haul" connection. I just have to get someone to dig the ditch and terminate the ends.

* I should be getting a home soon now that I can support myself with a full-time job. The home I am hoping to put a bid on has a nice basement and a separate large RV/semi-garage a few hundred feet down the driveway. As soon as I get the place, I plan to move all my systems out and into the new home.



The Obstacles:


1. Power - no real explanation needed for this one. Lots of computers will require lots of power. I'm not sure how to get around this. I've thought of solar, but that is a large investment to begin with for not that great of conversion efficiency. Then you have to concern yourself with cloudy days, batteries, and keeping the systems up at night. I've also considered wind, but the area I'm in doesn't have a lot of wind. The only other "alternative" energy I can think of getting diesel generators and converting them to biodiesel/vegetable oil. I'm aiming to do it as green as possible no matter what energy source I use (grid, solar, bio). Note that the primary objective here is to compute for science, astronomy, health, and climate change.

2. Heat - All those systems will produce heat. I can modify the building to have fans near the top of the wall, exhausting the heat that rises to the top. I've also thought of adding other fans at the top of the opposite wall to intake "cool" air. Air conditioning would be out of the question unless it was very cost effective and/or cost less to cool versus fans.

3. Time - I'm a full-time tech at a large corporation and part time college student. During the semester I don't have much time to devote to much other than classwork. My job is awesome, but I can't take off work suddenly if half of one of my clusters dies.

4. Money - Even though it might sound like I have a lot of spare cash, I don't really.



Why, you might ask?
Well, mainly, doing exactly this has been one of my goals for a long time now. I've wanted to find a way to put my computers and technical skills to use for science, health, astronomy, and the environment, even if it's just crunching numbers using the BOINC platform.

I plan to try to do this starting out small and build from there (testing the waters), whether it's completely ridiculous or not.

As mentioned, I had hoped to do it in a "green" way, to help out the environment as well. Speaking of which, I've gone as far as developing a small website laying out my thoughts and ideas to the public. The site is: www.greenpowerdatacenter.org
Unfortunately I have not got many hits on the site as I am unsure where to advertise.
I'm not trying to advertise my site here, but get it out there and also ask you guys if there is an appropriate place I could advertise it, or you knew of someone willing to advertise it.

I would like your thoughts on feasibility, advertising, project support, or anything else you would like to add.

Thanks for your time! Smile

 

[Jon855]

WOW... Quite bunch of computers you have there. In all likelihood, it's not exactly feasible but let me share you my thoughts.

- Donate some of the computers to needy organizations - This will cut down on clutter and make your project far more manageable.

- If you've a good shelter for the system, you may want to go naked "case-less" and obtain newer and better power supplies that you can bridge together for some bunch of the systems. (This should assist with the temperature and power distribution.) I would suggest putting them on open shelves so that the air can circulate.

- I say go with a router and distribute the connection by multiple switches - probably far more efficient that way than obtaining or converting a system into a router (you can if you want to use one as a router) but beyond that use switches and the router/WCG can be both configured to allow each device to communicate within specific time-frame.

- This sounds like a big project, if you don't mind my asking, where are you gonna be housing all of the rigs? (We need more DC Porn [hardware pics])

- I wish you the best on your journey, seriously donate some of them - you'll be doing a lot of good. And if it is not possible to be used - RECYCLE!

- And as for advertising a site - that's actually against the rule # 19.

(19) ADVERTISING, site pimping, contests or any type of business promotion is not permitted. Soliciting for fraternal organizations, humanitarian causes or personal rewards is prohibited unless approved prior to posting.

Get Kyle's permission first, most likely probably you'll be granted given that you are doing this not for your own benefit and for the greater good. But hey, Kyle's the man...

 

[alan2308]

With the cost it would take to run all those boxes for one month, you could pick up a couple GPU's which would outproduce the whole lot of them. They may be more valuable if running WCG, but I can't say what kind of production they'll do there. Without a free source of electricity, it's just not worth it.

As far as running OSCAR, it wouldn't be of any use for F@H or WCG. Its also a bit more complicated than installing from the OCSAR CD onto a bunch of boxes. Its not like Firefox becomes 50x faster by running it on a 50 node cluster. I would rule that out since you said you don't have time to put much into managing a large F@H farm.

 

[BillR]

I totally understand the frustration you must feel having all that total computing power and yet a quick cost benefit analysis will show they will never produce enough science to be viable.

If you had free power (no such thing really) maybe, but in the end it just doesn’t work. We have all hit that wall with gear we should have sold 2 years or more ago and yes, it hurts. Those who haven’t hit it yet will, blame Moore and the demanding needs of science.

“Alternative power” just isn’t here yet, it will be, just not yet.

What you might have is a tax break waiting to happen if you donate those machine; it all depends on your current financial situation.

Either way, some good thoughts

 

[Kendrak]

I say sell the systems on Ebay.

Take the cash and buy power efficient part that will do as much or more computing than the mass of systems.

I think you should be able to condence all of thoes systems down to a dozen or so, very specialized systems for the program of your choice (or both)

This takes care of networking, power, and heat problems.

 

[TC10284]

- This sounds like a big project, if you don't mind my asking, where are you gonna be housing all of the rigs? (We need more DC Porn [hardware pics])

I had planned to put them in the RV/semi building of the house I hope to purchase soon.

- I wish you the best on your journey, seriously donate some of them - you'll be doing a lot of good. And if it is not possible to be used - RECYCLE!

I intend to donate some of them! It's always a good thing to recycle and donate.

I think I am going to start small at first. I may not be able to get to the end result, but even if the science benefits are minimal, I'd be happy to know I'm helping somewhat.

You know I really appreciate all you guys' insight without telling me I'm a complete moron. It's why I like to hang around the Hard Forums. I don't have much time to post, but I like to read threads.

 

[TC10284]

You know, I just remembered that the Compaq Deskpro EN systems I have don't like running without a keyboard. I've found no obvious setting in the BIOS to enable keyboardless operation or disable "halt on keyboard error". I'm certainly not buying a bunch of PS/2 keyboards, or multiple KVM switches just because of that...

The IBM NetVista's do have this option.

That gives me a perfectly good reason to part the Compaqs out, sell, or donate them.

The downside is the NetVista's only have two 168 pin DIMM slots as opposed to three in the Compaqs.

Oh well, I suppose...

 

[Jon855]

You know I really appreciate all you guys' insight without telling me I'm a complete moron. It's why I like to hang around the Hard Forums. I don't have much time to post, but I like to read threads.

MORON...
















He had it coming

 

[Jon855]

Kidding aside, you are far better off by either donating it or selling ALL of them and get a few system - 10 or so and these ten system will most likely output far more work than these 250 system all combined - more work for less energy which mostly does fit in with your "vision" of greener computing when comparing to running 250 PCs for the same amount or less science.

Now - we just got to determine the "cost" for each system if that's possible. So we can further assist you with the project of yours. I'll be more than glad to provide a hand if needed. Just ask.

DC on!

 

[Pocatello]

I have been down this road before.... as have others here.

Here is my advice.

Take this one step at a time. You want to produce for science? Get one box folding for the cause you choose. Most of use here run F@H, but many of us run WCG and/or both at the same time. What I am saying is this... get one boxen up and running first. Don't buy ten PSU's, ten MBs, a load of memory... and let it sit because you don't have the other components yet.

What moronic thing did I do that causes me to give this advice? I went and bought some intel server dual socket motherboards. I ended up with ten or eleven of them. In just a short matter of time they became worthless. I never got a single one of them up and folding. I still have them... but I never got the CPUs for them. I also bought twenty brand new HSF's for them... but no CPU's.

Best of luck.

 

[alan2308]

Another thought on what to do with all those boxes. Have any broke friends/family that need a computer or a free upgrade to the one they have? Load F@H or WCG on one and give it to them under the condition that they leave the F@H or WCG service alone, and when they're done with it they have to give it back to you. I've got a couple boxes out in the wild like that.

That way you can still have some of them running but not have to foot the bill on all the electric use.

 

[Tigerbiten]

The standard client runs at around 50-75 PpDpGhz.
So 200 boxen @ ~1 Ghz will only produce ~10-15k PpD.
And even at 50 watts per box you looking at 10 kW.h

One of my Q6600 + dual 9800GTX+'s will give you around the same output with quick protiens and only pulls 350 watts ...........

It the main reason FLECOM no longer runs a big work borg.
Why have the hassle of running a farm of 500 computers for only 100k PpD when I can get the same output from 5.

Luck ............