I used the DSS Demo CD and the performance is the same. Sequential read with one worker is fast and with two workers the value goes down after a few seconds.
OK, so in my Enviroment there is in the current version FileIO the better choice.
Will there be changes in the future that more cache will be used in BlockIO? (When i'm right that the cachig in FileIO results the more performance in random IO.)
today I did update our iSCSI R3 Enterprise from 2.21 to 5.0 and did create new RAID and a volume with Block I/O. We use the iSCSI storage for our VMWare ESX 3.5 server virtual maschines. Inside a virtual maschine I use IOMeter to measure the disk performance and it was very weak. I was playing with the tuning options (InitialR2t = no, ImmediateData = yes) without performance boost. Now I deleted the Block I/O volume und made a new File I/O volume and get acceptable performance after playing with the tuning options.
With Block I/O I get 22 MB/s read and 5 MB/s write and with File I/O I get 57 MB/s read and 48 MB/s write (IOMeter is set to 100% sequential, 64k, 100% read or 100%write).