[rbridge] New shim header proposal---without F-tag field

[Gray, Eric]

Silvano,

	I did not actually say that we need to support half a
million RBridges, I simply said that we may need to support
more than 32K (or even 64K), consequently I do not support a
reduction in the "name space" to 16 bits.

	It would never work out to an order N^2 connectivity 
in any case, so this concern - IMO - comes under the heading
of FUD.  If every port needed to be connected to every other
port, then there would not be a need to have VLANs, and -
hence - no need to have separate per-VLAN instances.  In a
real world application where VLANs are used to connect large
numbers of multi-port switches, only a few ports on a few 
switches typically end up connected to each individual VLAN.
Some exceptions would occur, but they would be exceptions.
Thus per-VLAN connectivity would be a sparse partial mesh, or
VLANs would not be used.

	This is in fact part of the justification for having
per-VLAN IS-IS instances.  Trauma to part of the network is
only going to impact on routing convergence for the IS-IS
instances participating in affected VLANs.

--
Eric

--> -----Original Message-----
--> From: Silvano Gai [mailto:sgai at nuovasystems.com] 
--> Sent: Wednesday, November 01, 2006 11:55 AM
--> To: Russ White
--> Cc: Gray, Eric; Radia Perlman; rbridge at postel.org
--> Subject: RE: [rbridge] New shim header proposal---without 
--> F-tag field
--> 
--> 
--> What is a stretch is 32K trees of 32K nodes, which is what 
--> you need if
--> you want to deploy the current TRILL proposal with 32K RBridges.
--> 
--> Eric says that we need to accommodate half a million RBridges.
--> Half a million tree of half a million nodes is unfeasible.
--> 
--> If we want to support such a large number of Rbridges, we 
--> need to modify
--> the current Trill proposal, reducing the number of trees 
--> that need to be
--> computed.
--> 
--> Agreed?
--> 
--> -- Silvano
--> 
--> 
--> > -----Original Message-----
--> > From: Russ White [mailto:riw at cisco.com]
--> > Sent: Wednesday, November 01, 2006 8:35 AM
--> > To: Silvano Gai
--> > Cc: Gray, Eric; Radia Perlman; rbridge at postel.org
--> > Subject: Re: [rbridge] New shim header proposal---without 
--> F-tag field
--> > 
--> > -----BEGIN PGP SIGNED MESSAGE-----
--> > Hash: SHA1
--> > 
--> > 
--> > >> 	I don't think there is consensus yet that we 
--> only need 16 bits
--> > >> for RBridge IDs.
--> > >
--> > > With 16 bits, using 1 bit to indicate unicast, we can have 32K
--> switches.
--> > > According to the current definition of TRILL, there is 
--> the need to
--> run
--> > > Djikstra on a database with 32K records, one time for the core
--> instance
--> > > and 32K times for the IRTs. This is clearly out of 
--> reach even for
--> the
--> > > most powerful CPU.
--> > 
--> > ?? Perhaps 32k trees is a stretch, but 32k nodes in a 
--> tree? I don't
--> > think that's really a stretch on today's processors, from 
--> the scaling
--> > work I've seen done in IS-IS.
--> > 
--> > :-)
--> > 
--> > Russ
--> > 
--> > - --
--> > riw at cisco.com CCIE <>< Grace Alone
--> > 
--> > -----BEGIN PGP SIGNATURE-----
--> > Version: GnuPG v1.4.2.2 (MingW32)
--> > Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
--> > 
--> > iD8DBQFFSMypER27sUhU9OQRAqLwAJ9FLN2YNP2mrB6i0ZcV1fxLzAT0hgCfYHYj
--> > 6mJCA68nQLjCD8US9nRZZag=
--> > =kwzq
--> > -----END PGP SIGNATURE-----
-->