Re: iscsi single control channel

[Mike Kazar <kazar@icubed.com>]

At a high level, I like this proposal a lot.  The feature that makes me 
most uncomfortable about iSCSI as it exists is this: iSCSI uses multiple 
TCP connections, partly to get good performance, but then requires an 
*additional* windowing protocol layered on top of multiple TCP connections, 
sharing state between those connections, to reimpose the desired ordering 
constraints.

This means that even if you throw hardware at the problem of running TCP 
well at high speeds, you still need independent hardware to handle iSCSI 
sequence number processing at high(er) speeds.  This iSCSI hardware will be 
more expensive, since it will only sell to a smaller market than a general 
TCP accelerator.

Kalman's proposal has the benefit that requests can be removed from each 
connection and processed independently, without any further 
protocol-imposed synchronization constraints.

	Mike Kazar
		(kazar@spinnakernet.com)