Re: iSCSI Autosense Consensus, Connection next steps

["John Hufferd/San Jose/IBM" <hufferd@us.ibm.com>]

Steph,
I certainly agree with your first point.  However, we need to focus a
little more on the second point.

You talked about Transport layer and Wedge layers.  I guess I do not know
exactly what to call iSCSI if given these two options, other then to
consider iSCSI part of the SCSI Transport (layer).  It is a device specific
driver, which in this case deals with iSCSI Devices.
I do not think we should call iSCSI a Wedge Driver.  In fact if you do not
have iSCSI supporting the balancing of commands and data across the
multiple connections (and perhaps some generic Alternate Path Retry), you
will have to have a Wedge Driver Do it.  In was David Black's point when he
said that Wedge Drivers are needed anyway to handle a number of vendor
specific functions.

I think that David is correct the only point that I was making was that if
iSCSI was able to do the load balancing and some Generic Alternate Path
recovery, that the Vendor Specific Wedge Drivers will be much simpler.

Having said all that, I think the point you wanted to make was: if the
command load balancing functions can not be done in the xxxx/IP transport
then no one should do it.  That is clearly NOT an option, since the various
Wedge Drivers do that today and will do that in the future.  So the
remaining point is: considering only the iSCSI & xxxx/IP layers, do we want
load balancing to be done only in the xxxx/IP layer or can it be done in
the iSCSI layer.  The iSCSI Device Driver might be easier to create, if the
xxxx/IP layer did the Balancing.  But, I think a number of vendors of iSCSI
will be willing to do this balancing if the xxxx/IP does not do it.

 As long as it is not required for every one to build multiple connections
per session (remember: the default in Synchronous is a single connection
per Session) then Synchronous permits those that want to build a balancing
and generic Alternate Path Retry function, to do so.

.
.
.
John L. Hufferd



Stephen Bailey <steph@cs.uchicago.edu>@ece.cmu.edu on 09/05/2000 02:44:21
PM

Sent by:  owner-ips@ece.cmu.edu


To:   ips@ece.cmu.edu
cc:
Subject:  Re: iSCSI Autosense Consensus, Connection next steps



> Second, on connections, I haven't seen enough discussion to call
> consensus, but I am going to try to narrow the option space and
> structure the discussion.  Four models for sessions have been
> proposed:
>
> (1) Symmetric - all connections usable for command and data.
> (2) Asymmetric - single command connection, others are data.
> (3) Split - assign LUN sets to specific connections or pools of
>    connections.
> (4) SCTP - use SCTP's support for multiple connections.

For the purposes of shaping the consensus, here's my stance:

  1) SCTP should only be pursued if TCP does not admit a viable
     solution to the iSCSI requirements.


Given that there are many iSCSI community participants expressing the
belief that iSCSI on TCP IS possible, I believe iSCSI on SCTP
proponents are just going to be stuck holding an `I told you so'
card.

This gates my second statement:

  2) multiple connections per session should only be supported if the
     underlying transport (e.g. SCTP) layer supports it.

Obviously, TCP does not presently support this abstraction, so
assuming SCTP gets killed for now, I am against multiple connections
per session.

In general, I am dubious that connection bundling above the transport
layer, but below some more application-informed layer (e.g. a wedge
driver) will work acceptably.  However, if the transport layer
provides it, then, by definition, it must work (ha, ha), or at least
it's not iSCSI's place to say that it won't.

Steph