Hi,

we are planning to set up a linux cluster a my work. Right now we are in the process of setting up a proposal for this and we thought it could be helpful with some expert opinions.

About 5 people will use the cluser and various types of programs will be run on it. Typically each run takes about a week on a single processor machine. With a cluster we would run longer jobs, though.

What type of hardware would you recommend (which bottle necks should be avoided)? Number of processors? Network requirements? Software?

Thanks in advance, Johan

Please look also to the STAR-CD forum.

What comes first is the nature of software you wish to run and the general size of the machine. This will determine whether you should go for fast (expensive) interconnects with relatively few processors but of the fastest kind (expensive) or slow interconnects (cheap) and lots of cheap processors. For example, heavily implicit methods are going to nudge you towards the former whereas explicit/particle methods towards the latter.

If you are unsure, find out what solvers the codes use and look at sets of results for the NAS Parallel Benchmarks which cover a range of solvers and a range of grid sizes typically used in CFD. This should help you gauge whether relatively cheap ethernet (fast or the more expensive gigabit) is going to be up to the job or whether it is necessary to go for the expensive connections such as sci/myrinet/...

If you end up going for the cheaper nodes then performance details of the ethernet chips, processor memory systems, interconnection software and such are going to be important and signficantly influence overall performance. It can be quite difficult accessing reliable information on this topic. If you go for expensive nodes then this tends to be largely sorted by the supplier of the fast interconnects + connection software.