The following stats show reading off 2×1.5TB TB disks and 2x60GB SSDs.
The graphs show when my system peaked in IO Rate and that the usage was far too high for the little dualcore box. Mostly, I suspect, not because it’s unable to handle these IO rates, but rather because the 4 “dd” threads clobbered the CPU (and …locks). When I pushed 600MB/s with just the SSDs the system was still very responsive.
Lessons learned… :
Redo the test using 4 Solid State Disks, see if this scales to above 1GB/s and see how the system usage evolves. On the larger hosts I would be able to dedicate 1-2 cores to Xen “overhead” and pushing IO, but more would be undesirable. On the other hand scaling out is pointless if I won’t see a bandwith of above 400MB/s per Xen host.