Beuwolf Mini-supercomputer Cluster

aphexcoil

aphexcoil

I am gathering more information on the computer cluster and I have determined that I can build a mini-supercomputer for less than $10,000.

Here is the configuration of each of the client machines that I am pondering:

Motherboard: ASUS P2PE Intel 845e Chipset ($120)
CPU: Intel 3.08ghz Pentium IV ($250)
Memory: 2700 533mhz 1 Gig Modules ($150)
Network: FiberOptic 1gigahertz ($150)
Hard-Drive: 20 Gigabyte ($40)
Case: $50

Total cost for client machine: approx. $750

10 machines would run around $7,500

I don't need video-cards since the motherboard has in intergrated one. I also don't need monitors since I can log into them from the server.

Linux is free.
 
Quote from aphexcoil:

I am gathering more information on the computer cluster and I have determined that I can build a mini-supercomputer for less than $10,000.

Here is the configuration of each of the client machines that I am pondering:

Motherboard: ASUS P2PE Intel 845e Chipset ($120)
CPU: Intel 3.08ghz Pentium IV ($250)
Memory: 2700 533mhz 1 Gig Modules ($150)
Network: FiberOptic 1gigahertz ($150)
Hard-Drive: 20 Gigabyte ($40)
Case: $50

Total cost for client machine: approx. $750

10 machines would run around $7,500

I don't need video-cards since the motherboard has in intergrated one. I also don't need monitors since I can log into them from the server.

Linux is free.
More...
You will need a switch that can do Gigabit ethernet - add about $1K for a managed one, which I highly recommend. I would also add a good KVM switch. The Tripplite that I use is about $300, but only supports eight nodes. However, I believe they can be stacked as your needs grow. Don't forget you need to spend about $15 - $20 per cable set each for the KVM.

I doubt what you are getting is Fiberoptic, because if it were, you would be spending a fortune on getting an all fibre network. The switch alone would be several grand.

Gigabit ethernet is not bad and is relatively inexpensive and will work fine for what you are asking it to do.

As you may be aware, this is a fraction of the costs. What about develoment costs? What about datafeed costs? IF, and it is a BIG if, you are able to spot innefficiencies as a result of your cluster, you will find that the one fastest to the exchange is the winner. Then the real expenses begin...

Here are what the pros use for fast interconnects on clusters:

http://www.myricom.com/
http://www.dolphinics.com/
http://www.scizzl.com/
http://www.quadrics.com/

There are also things that you can do that you haven't even imagined yet that give you the extra speed edge where it counts, but that is in a Galaxy far far away from where you stand now...

nitro
 
Quote from aphexcoil:

I am gathering more information on the computer cluster and I have determined that I can build a mini-supercomputer for less than $10,000.

Here is the configuration of each of the client machines that I am pondering:

Motherboard: ASUS P2PE Intel 845e Chipset ($120)
CPU: Intel 3.08ghz Pentium IV ($250)
Memory: 2700 533mhz 1 Gig Modules ($150)
Network: FiberOptic 1gigahertz ($150)
Hard-Drive: 20 Gigabyte ($40)
Case: $50

Total cost for client machine: approx. $750

10 machines would run around $7,500

I don't need video-cards since the motherboard has in intergrated one. I also don't need monitors since I can log into them from the server.

Linux is free.
More...


I seriously question the pricing you have put up.

One Gig PC 2700 modules for $150.00?
3.08ghz Pentium IV for $250.00?

I don't think so.....

Crucial Ram Pricing

CPU Prices
 
Quote from nitro:


You will need a switch that can do Gigabit ethernet - add about $1K for a managed one, which I highly recommend. I would also add a good KVM switch. The Tripplite that I use is about $300, but only supports eight nodes. Plus you need to spend about $15 - $20 per cable set each.
More...

That is good information -- thanks! Part of the process for me is to accumulate as much knowledge as possible, then start cracking equations and see if I can observe any sort of inefficiency that I could profit from -- and if I do see one, then I can start this project.

I'll need to find someone who is more knowledgable than I on computers, though -- since some of this is reaching my limits on what I know and don't know.

One day at a time ...
 
Quote from aphexcoil:



That is good information -- thanks! Part of the process for me is to accumulate as much knowledge as possible, then start cracking equations and see if I can observe any sort of inefficiency that I could profit from -- and if I do see one, then I can start this project.

I'll need to find someone who is more knowledgable than I on computers, though -- since some of this is reaching my limits on what I know and don't know.

One day at a time ...
More...
Not meant in any offense, but it is because you are so naive that I encourage you to try. If you knew what I knew, you would not attempt it without the aid of some very deep pockets and a solid business plan...

BTW, you better get started learning MPI and PVM:

http://www-unix.mcs.anl.gov/mpi/
http://www.csm.ornl.gov/pvm/pvm_home.html

nitro
 
Build a smaller cluster until you get it to work. By the time you get it up, it will be obsolete. These are scalable; start with 3 or 5 machines.
 
You must log in or register to reply here.
Back
Top