But using File I/O with plenty of RAM and fast enough CPU (Xeon QC) should benefit from the filesystem caching to improve read speed? So I don't understand how Block I/O would be faster than File I/O in a virtualized environment like ESX or XenServer? We are currently building two systems (one failover) with 5 SAS 15k and 5 SATA disks, of which both arrays will be an iSCSI target for XenServer hosts. Until now, we were convinced File I/O is the way to go. Could you explain a little bit more what new insights you have on this, since before File I/O was recommended?

Thank you.