I have two Dell PowerEdge R610's which I want to use for numerical computing. Each R610 has 6 harddrive slots, 2 of which I will put under Raid 1 (for local OS and local data IO). What performance difference (for both, IO and CPU) can I expect with the following alternative storage configurations:

1) Putting the 8 (spare) harddrive slots (4 for each R610) under Raid 5 and using them for central cluster storage (i.e. /opt) -- the number of cluster computing nodes will grow in time.
2) Using a dedicated central storage system (e.g. HP Eva with 8 harddrive slots) for reliable storage.

Thanks.

The Dell systems use PERC (Power Edge Raid Controller) cards for the most part and these are OEM'ed from LSI. While they perform well in general you'll can get better performance from SAN attached (fibre) disk arrays especially if you use redundant paths (and Linux multipath). In our environment we generally use SAN attached disk arrays for performance but rely on the PERC for the OS and other filesystems that may not need as much performance. The other benefit to SAN attached storage is it can easily be made available to a separate server if your primary dies for some reason.

The question would be how much performance you really need and the IOPS allowed by each of the configurations. I've never really delved into what the PERCs atually allow for IOPS because we have large disk arrays so our default is always to SAN attach.