Re: iSCSI: Canonical Targets

[Mark Bakke <mbakke@cisco.com>]

Stephen Bailey wrote:
> 
> Mark,
> 
> I don't disagree with the general thrust of your statement, but I must
> argue with:
> 
> > Even in an environment without proxies and gateways, any
> > medium-sized disk array should be able to support more than one

I did say "should".  OK, perhaps I'm engaging in a little bit of
wishful thinking, but it seems to me that most people would rather
deal with a set of logical targets, each created for use by a host
or application, rather than a single target with different LUN maps
depending on who's asking.  I know we'll have to support the latter
as well.

Anyway, I agree that bridges will always need to take up the
slack, but my point was that I didn't want to see initiators
make the assumption that there is only one target for them at
each address.  I just don't want to see us do a repeat of some
of the limitations of ||SCSI and FC.

> > logical target.
> 
> Unless somebody's really building a medium sized disk array a)
> completely from scratch b) for iSCSI use only, this is unlikely.
> 
> Existing disk arrays present all storage containers as LUNs.  The most
> likely [credible] iSCSI targets will be existing storage arrays with
> iSCSI front-ends.  This follows the ||SCSI->FCP pattern.  The
> alternative interpretation, which I'm pretty sure is false, is that
> the back-end part of a storage controller is really simple so the
> serious iSCSI targets will not come from the same vendors (or groups)
> as existing, serious storage targets.
> 
> Bridges are a different issue, but I've always felt (and you seem to
> agree) that the bridges just have to suck it up and, well, bridge, the
> semantic gaps between what's expected in a native initiator->target
> interaction and what you get with an initiator->bridge->target one.
> 
> Steph

-- 
Mark A. Bakke
Cisco Systems
mbakke@cisco.com
763.398.1054