I did a comparison of cloud HPC approaches including Amazon ECC here -

http://www.ibm.com/developerworks/cloud/library/cl-ind-cloud-e-learning2/ and

http://www.ibm.com/developerworks/cloud/library/cl-cloudscaling1-hpcondemand/

I think you can build a modest cluster of Linux machines yourself and get more out of it if you plan to use for a good length of time (say 2+ years). Some options to consider include GPU co-processing and use of CUDA and/or OpenCL as well as distributed computing tools (noted in the paper). CentOS is a great choice, but for ease of installation and use I like Ubuntu 12.04 LTS myself - it's just easier to install and configure with tools like CUDA, OpenCL, etc.

I suspect your workload would benefit from a vector co-processor and CUDA as well as GPUs for rendering. However, if you just need lots of cores, another option that may be of interest is MICA (the Intel Xeon Phi co-processor).

The advantage of AWS EC2 and other pay as you go HPC options like Penguincomputing (http://www.penguincomputing.com/services/hpc-cloud ) is that you can solve a problem and not own the equipment, but again, if you in this for the long haul, that's perhaps also the disadvantage - the idea of HPC on-demand is an interesting compromise you might consider - build your own small cluster and then use Cloud HPC for really big runs - this is my favorite approach. I find that I need to develop and test on my own small installation anyway (or it goes best if I do) and then use university resources or Cloud HPC once I'm ready for production runs.

As far as storage, Linux software RAID is an option along with Ceph if you're willing to pull together a JBOD and do the SAN/NAS configuration yourself.

If you build your own small scale system, I find it helps you make better use of Cloud HPC since you'll know better what to ask for and look for from any HPC on-demand service. You might find your private system is good enough.

Hope that helps.

(1-2)solaris or federal or redhat linux as edition etc. are all ok. and there is a special windows os for the cluser named Windows Compute Cluster Server.

(3-4)there are some hardware provider can do this , the hardware and software are a whole package. like lenove hp and ibm

Solaris is suitable for this as well windows launch new os along with windows 8.1 server for the high computing cluster while earlier one was windows server 2000 but you don't need to bother about all the things because the hardware provider suggest you according the computing limit setup at the time of installation.

Hope this ill help.

CentOS is recompiled Redhat Linux.

We use that here, and run WRF on such systems as well as CFD and

Computational Chemistry and BioInformatics.

The OS really is not important as the scheduler. 

You must use a job scheduler to dedicate nodes, otherwise concurrent runs will interfere with one another.  The job (sometimes called batch) scheduler will optimize system throughput and will also mimimize average single job completion for jobs of about the same size.

We use Torque/Maui as the scheduler.

Why you need a batch or job scheduler:

Suppose you have two MPI jobs to run on two compute nodes that use all the resources but only half the memory of each node. Assume they complete in 24 hour when run alone..

If you launch both at the same time, then both will run at 50% speed, so you will get both answers after 48 hours.

If you launch one and let the second wait in a queue for the first to complete, both run at 100% speed, so the first gets done in 24 hours and the second in 48 hours, no slower than if both were run at the same time.

So in using the scheduler, the worst case is no worse, and the average runtime is 36 hours, which is better than running both at the same time.

Also note that it required twice the memory to run both at the same time, so you had to spend more money to get worse average performance.  If you did nothave twice the memory, you would get page memory thrashing and nothing would get done.

Thanks for the information. Indeed it is the batch scheduling software - especially in the context of a HPCC that I am trying to figure out. I have recently seen a smallish custom built HPCC where centos is separately installed on every one of the compute nodes.

You can use the free torque scheduler, and for a small cluster just use its 

pbs_sched, and skip maui. 

You can ask questions on the mailing list.  Sign up at:

http://www.supercluster.org/mailman/listinfo/torqueusers

You can download and get the administrators guide for torque at:

http://www.adaptivecomputing.com/products/open-source/torque/

Thanks for your answers. In the end our tender evaluation compared Huawei and HP clusters with a SGI UV3000 monolithic machine, all at similar prices of about 550k. We went with the SGI for a variety of reasons, local support being one of them, and run RHEL and are experimenting with the torque scheduler.