RE: iSCSI: CmdSN and Retry

["Y P Cheng" <ycheng@advansys.com>]

Doug,

I think you have misunderstood what I said about error recovery in different
layers.   I specifically have stated that we should separate the three
layers: 1) Transport layer provides in order delivery on Internet, 2) iSCSI
keeps the in order among multiple connections, and 3) SCSI has its own in
order command semantics.  Frame retransmit and congestion control belong to
the transport.  Digest error retransmit belongs to iSCSI layer.  Finally,
SCSI itself has done command level recovery very well for the last 20 years.
We should mix iSCSI and SCSI retries.  I want to be able to pipe 100,000
frames down the wire without letting retry and ACKs getting into the way.
In my previous writings, I have stated that for TCP to support 10Gb/100Ms
network with a huge reassembly buffer, message framing and TCP/RDMA must be
used.  Whether the maximum TCP bandwidth supports such a high IO rate is
beyond the Charters of this WG.  I have no problem in using SCTP, if this is
what it would take for the 10Gb/100Ms network.

Y.P.

> Any TCP bandwidth is limited to .93 MSS/(RTT * loss_rate^-2) so your
> assumption of one in a million at 100 milli-seconds limits at
> Fast Ethernet
> rates.  There will be significant overhead on long pipes dealing with TCP
> handshakes and any additional handshake will not add significantly to this
> resource overhead.  The more immediate these handshakes, the less
> resources
> consumed with respect to any needed replays to satisfy the
> transport.  I do
> not agree procrastination that depends on rediscovery of indexes unrelated
> to these handshakes to be more effective in allowing pipe-line execution.
> Unless the transport maintains order, there will be a requirement to wait
> for each command to complete to assure the sequence.  This makes
> compliance
> to SCSI impossible and not a means of improving the pipe-line.
> An immediate
> handshake within the transport layer dealing with digest errors
> is the best
> means of improving performance.  Reliance on the SCSI tag to
> recover from a
> failure deduced from a dropped transport sequence is not a clean
> separation
> of layers.  The amount of resources saved by deleting this positive
> relationship is dwarfed by the window size required for such a fat pipe.
>
> Doug