Re: iSCSI: New iSCSI MIB draft

[Keith McCloghrie <kzm@cisco.com>]

Hi Bill,

(I'll post your message and my response to the list as you said I could.)

Yes, a socket interface typically provides the ability to listen
for incoming connection requests on one or all addresses.  That is,
a common implementation of the IP-layer allows for what iSCSI is
proposing.  However, few if any applications do it.
So, just because a lower-layer implementation allows for it to be
done, should the application's arhcitecture require it be done ?

Binding to an interface does seem better, but I think it causes iSCSI
to enter the murky world of multi-homing, see the descriptions of the
Strong ES model and the Weak ES model in RFC 1122 (pages 62-63).
(Or perhaps, iSCSI is already in that world?)

Keith.

> Keith,
> 
> I want to pick a nit with your description of how a TCP/UDP Server works...
> I'll do this privately, if you think it is worth expanding on feel free to
> post to the IPS mailing list
> 
> Comments inline
> 
> 
> -----Original Message-----
> From: owner-ips@ece.cmu.edu [mailto:owner-ips@ece.cmu.edu]On Behalf Of
> Keith McCloghrie
> Sent: Wednesday, November 07, 2001 8:27 AM
> To: Mark Bakke
> Cc: Michele Hallak - Stamler; IPS
> Subject: Re: iSCSI: New iSCSI MIB draft
> 
> None of the IP-based applications that I can think of (e.g., FTP, HTTP,
> SNMP, SMTP, BGP, etc.) have a static binding of a local TCP port number
> to a subset of the local host's IP-addresses.  Rather, they have a
> static TCP/UDP port on which the server listens, and the client has a
> port dynamically assigned on a per-connection/session basis.  The
> server listens on the static port on all local IP addresses; the client
> dynamically picks a local IP address on a per-connection/session
> basis.  So, does iSCSI work this way also ??  If so, then the MIB
> is wrong; if not, I think iSCSI has a problem with DHCP.
> 
> <Bill Strahm>
> Actually the server may bind to ANY local address, in fact in many
> situations you want the service to bind to one interface but not the other
> (let say providing DHCP internally, but not externally, or providing
> internal DNS, vs. external DNS).  This binding is done based on IP address
> (the parameter provided to the bind call), you are correct most times a
> default address is provided as 0.0.0.0 which means bind to all interfaces.
> 
> What you want to do (and I think will solve the DHCP problem) is to bind to
> an InterfaceIndex, allowing the application to determine which IP address to
> use based on entries in the IP table (what interface is that address bound
> too)
> <End Bill Strahm>
> 
> Also note that you have used InetAddressType (from RFC 2851), and one
> of the enumerated values of InetAddressType is 'dns'.  As and when 'dns'
> is the value of iscsiTgtPortalAddrType or iscsiIntrPortalAddrType, then
> you would typically not have a static binding of a local TCP port
> number to one of the local host's IP-addresses.
> <Bill Strahm>
> I think you can still statically bind to a single interface in this case
> <End Bill Strahm>
> 
> Bill Strahm
> +========+=========+=========+=========+=========+=========+=========+
> Bill Strahm     Software Development is a race between Programmers
> Member of the   trying to build bigger and better idiot proof software
> Technical Staff and the Universe trying to produce bigger and better
> bill@sanera.net idiots.
> (503) 601-0263  So far the Universe is winning --- Rich Cook
> 
> 
>