I've been doing some benchmarking on our filer (f540, dot4.3r4 256+8, 26 4g discs) and I'm trying to work out how things will scale on more recent filers. For our workload the filer pegs out doing around 1k nfs ops. This is slightly under half the sfs1 results. Is it fair to assume that given that data point that we can just scale the current models sfs2 results down by 1.25 to get the right sort of numbers? I know that sfs1/2 results are different but they do seem to be in the same ballpark. Workload is lots of small random reads which to me says number of spindles is going to be the critical factor which probably won't scale.

Chris