There are a couple of things here:
1) StarWind itself has extremely effective write-back or write-thru caching. As we deal with a very limited number of I/O scenarios we have huge benefit of optimizing our cache entirely for them (unlike Windows Cache Manager we don't need to be generic).
2) With SAN you're double buffered. I/O is buffered at initiator side (Windows SMB RDX is actually performance pig...) and at target side as well. So pipeline between two nodes is fully working. With SMB it's not like SAN a) load pipeline b) send many packets c) unload pipeline rather it's a) load pipeline b) send a couple of packets c) unload pipeline d) load pipeline e) send a couple of packets ... ... ... With SAN TCP stack is working at the full speed, no "pulsating" network file system traffic.
3) SAN concept is in general faster then any redirector (network file system).
That's all...
dsimmons@engsin.com wrote:I was wondering why writing to an starwind iscsi target is 10x faster than writing to the same windows volume that has the img file for said target. I am not complaining, this is freakin awesome. I copied a file from a server to the f$ share on my starwind box at 127megabytes per second. When I copied that same file from a server to its isci volume -- which is an HA device store on the F drive of my 2 starwind servers, the rate went up to 1.2gigabytes/sec!! Is SMB just that much more inefficient than iscsi?
anyway, just wondering.. Me and the other IT guys here were stunned.
