Maybe I am not fully versed on things, but something doesnt make sense. Lets say you have an ISCSI san with a 1gbe uplink. You have an ESXI server connected also via 1g. Wouldnt it be very very easy for that 1g to be the bottleneck, even if using just one or a couple SATA drives, since ive seen sata drives that can hit 100MB/sec(which would come close to 1gb/sec). if thats true, then how would adding a bunch of spindles help performance if you are always stuck at 1g?
I am looking to experiment with mpio. My setup will be one ESXI box(dual gig nics just for storage) and an openfiler box with 2 gig nics. the nics from the openfiler box will be directly connected to the ESXI box. Would the best way to achive higher than 1g throughput be to setup MPIO on the esxi box, then on the openfiler box open the target to both NICS, giving them both different IPs, or is there something else/additional I would have to do
This will be with OF 2.3, and esxi 4.1 trial so everything should be unlocked.
If you're doing something like video streaming then you could potentially saturate the link with fewer drivers (although you would also have to be able to process near to 125 MBps (and just not push that from the iSCSI server to ESXi). In reality most I/O tends to be random in nature and the seek time for a drive really slows down how much data can be retrieved.
I haven't used Openfiler much. Are you able to configure two targets for it? The iSCSI configuration guide can walk you through the setup for ESXi. Essentially you'll create 2 vmkernel ports for the vSwitch that uses your 2 storage NICs and then bind those to the iSCSI initiator. You'd then put in both target addresses when doing iSCSI discovery.