Re: SCSI URL scheme [WAS: Re: iSCSI: 2.2.6. Naming & mapping]

[Daniel Smith <dfsmith@almaden.ibm.com>]

Douglas Otis wrote:
> 
> You base this comment assuming there is no other means.  There is no reason
...

I aplogize.  I'm afraid that I do not know how to respond to your questions
and suggestions.  I politely request that you propose a naming scheme
that meets the minimum requirements, and also specify your assumptions, which
seem to differ radically from mine.  Then submit the proposal to peer
review.

The iSCSI Internet-Draft requires a text string to identify a service
delivery port.  My proposal (which is just a proposal, and was mulled over
while the iSCSI draft was being written, with the knowledge of the design
team) treats these strings as URLs.  It is designed to cope with and blend
the storage world with the network world.  The Internet especially presents
some challenges to the traditional storage notion of addressing.  I do not
pretend that the solution is a simple one, but neither is the problem
simple.  But iSCSI (the transport) is distinct from iSCSI (the URL naming
scheme), at least until everyone agrees that it is the greatest, most
frabulous, brabulous, zip-zoop dabulous way of naming stuff.


My starting assumptions (broadly and somewhat imcompletely) are,

a) The inter-network is controlled by someone who isn't me.

b) I can make a network of my own that I do control, but that can't talk to
other networks.

c) The Internet is named using DNS, which maps to ephemeral IP addresses. 
The DNS is moderately, though not totally, secure.

d) Firewalls are everywhere.  On your outgoing connection and on the second
party's incoming connection.  Possibly in the middle as well.

e) The network world has sufficiently many nuances/protocols/transports that
no protocol can totally manage a discovery mechanism within the Internet. 
This means the naming must be flexible, and the (iSCSI) transport separable
from the naming.

f) No matter how much I scream and shout, the network administrators still
aren't going to change infrastructure that already works for everyone else.

g) The network infrastructure will gradually change after iSCSI becomes a
noticeable consumer of bandwidth.

h) SSPs are going to appear.  (Have appeared?)  So people are going to be
running SCSI over the Internet before the network folks even notice.

i) 3rd party commands will become more important as the SAN network
increases in size.  (Actually, this one is dubious.  Let's just assume that
some people regard 3-P-Cs of critical importance.)

j) Any packet going over the Internet can have is source IP, destination IP
and contents sniffed out by an observer.  Hence all DNS names can be
discovered by inspecting and deresolving the traffic.

k) People using SSPs will be paranoid about security.

l) SSPs will have clients numbering in the tens of thousands, and storage in
the petabyte range.  All going through one entry DNS name.

m) Who was it who said that something good should be as simple as possible,
and no simpler.  (Einstein, Twain?)

There are, of course, many more assumptions.  These seemed the most
pertinent.

I am going to go into silent mode for a while.  Just to see what happens,
and if another contender for naming appears.

Daniel Smith.
-- 
IBM Almaden Research Center, 650 Harry Road, San Jose, CA 95120-6099, USA
K65B/C2 Phone: +1(408)927-2072 Fax: +1(408)927-3010