• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Spec me a small cluster...

Duff-Man

Duff-Man

Duff-Man

Soldato
Joined
24 Jun 2004
Posts
10,977
Location
Manchester
I have between £5k and £10k to spend on a small cluster for use by our office, if we can get something worthwhile for that money.

I've got plenty of experience with speccing and building PCs, but I've not got any experience with parallel clusters. So, if anyone could point me towards the equipment I need then that would be great.

I was thinking of something based on the core2 architecture, rather than i7, in order to keep costs down. Also, the speed of the interconnects isn't so vital for the application we have in mind. About 8Gb memory per CPU would be ideal. We should have plenty of rackmount space free.

Any suggetions or good links would be appreciated :)
 
What is the cluster for and how many machines were you thinking? e.g. are we talking about a resillient cluster where you only have two machines (running either in active/active or active/passive) or a cluster for rendering (for example)?
 
Hades said:
What is the cluster for and how many machines were you thinking? e.g. are we talking about a resillient cluster where you only have two machines (running either in active/active or active/passive) or a cluster for rendering (for example)?
Click to expand...

The cluster would be used for running large-scale scientific models. Numerical simulations of fluid flow and plastic / elastic deformation etc. A lot of the runs will just be in serial, so the cluster will be somewhere to dump runs that might take a couple of days, to avoid wasting performance on our desktop machines. We will also be running small parallel jobs (probably up to about 8 threads).

Big jobs will be sent to the main University cluster, but that thing is at about 50% downtime, so it can be frustrating (hence the reason we want to buy a small local cluster if possible). The University is happy to spend £2m on a new central machine, but they won't fork out £30k a year to hire someone to look after it. Typical big institution false economy.

Anyway, thanks in advance for any advice you can give :)
 
Well I'm still not really sure on how you intend to use the cluster or how jobs will get dumped onto it (ie whether users just have a choice of single machines to drop work onto or whether they want to see a single entity (which is actually a cluster and shares the workload across all machines). But it doesn't sound like you need high resilliance for failover in the event of losing one of the systems. So that makes things a bit simpler.

But generally speaking you will need:

A fast network between the machines.
A fast disk subsystem if there is lots of data being read or written.
Resilliant storage - look for RAID edition drives which are also certified for 24/7 use.
Lots and lots and lots of RAM.
A fast CPU in each machine.
A UPS (get a high end one so it can run for a while with no mains power).
Hotswap drive bays so a faulty one can be swapped out without bringing the system down.
A case with redundant PSU's so one can be swapped out without bringing the system down.
A case with good cooling (and a HSF to match) - Keep it cool so it stays reliable.


Personally I'd be looking at a high end i7, the maximum high end RAM that the X58 motherboard will support (24gb I think), a dedicated hardware RAID controller card running RAID 0+1 for speed and safety and a gigabit network switch connecting them up.
 
Last edited:
You must log in or register to reply here.
Back
Top Bottom