One more question: I was trying this free DSS Lite last month, checked the iSCSI part and got good performance (IOMeter inside virtual maschine). Does DSS Lite use Block I/O by default and if yes, why it was so fast and iSCSI R3 Enterprise is so slow?
Hard to say. I was using DSS Lite on another maschine, same board and NICs but older processor and less RAM. I did not configure much and I'm not sure if I created a RAID. I just had this DSS Lite to find out best settings inside the tuning options. Can RAID be the problem? With the iSCSI R3 Enterprise I use software RAID 5 on 4 harddisks, left-symmetric, 64k chunk size.
Thanks for your feedbacks. I was not using hardware RAID on DSS Lite or iSCSI R3. On iSCSI R3 I use software RAID 5 and on DSS Lite I'm not sure what I did configure. But I know I used default Block IO with DSS Lite and got good performance while testing with one VM and IOMeter. Testing few virtual disks with IOMeter I never did. Now I cannot play with my DSS Lite because the activation is gone and I dont want to start the activation procedure with putting the maschine in front of our firewall...
When I use Block IO with iSCSI R3, I got no message box telling me that File IO is suggested with VMWare.
I will use File IO with my iSCSI R3 and give up playing with Block IO.
What I wish is more support from Open-E giving us information what are the best settings using their products with VMWare. I was reading Block IO is 30% faster than File IO. So I moved all our VMs to another storage, made the update, recreated a volume with Block IO just to realize the performance is pretty poor and File IO is recommended for use with VMWare.
Radui Wrote:
> BlockIO is only fast when a single task is working on the disks.
We found this to be incorrect. We use iSCSI target as a Virtual Tape Libray (VTL). The data in the VTL is also backed up to tape for offsite storage. This is single access, long sequential read, when being offloaded to physical tape and the peformance of file I/O is double that of block I/O. Block I/O gives no more than 30MB/sec, where as File I/O we get 60MB/sec.
We have done extensive testing with IO Meter and file I/O wins everytime. I don't understand why block I/O is the default when it's clearly slower (at least in our case anyway).