[rbridge] IS-IS per VLAN ?
touch at ISI.EDU
Tue Sep 27 07:29:58 PDT 2005
Guillermo Ibáñez wrote:
> Joe Touch wrote:
> Dino Farinacci wrote:
> I would expect that a set of rbridge nodes could work anywhere a single
> bridge could work, but that the overall total number of devices on the
> L2 served by the system would not change substantially.
> I.e., the primary goal of rbridge is better bandwidth between nodes by
> the use of an alternative to a spanning tree inside the rbridge campus,
> but it doesn't solve the problem that it's still part of the L2 system
> into which it is plugged, and so I wouldn't expect scale to change much
> - EXCEPT where it is limited by spanning tree convergence.
>> I am not sure I understood well the argument provided by Joe on keeping
>> the actual scale, so I apologize in advance if I missed the point.
>> It seems a bit conservative to assume that the current L2 network size
>> will not change.
It's not as much conservative as it is focus of the architecture or an
>> The limitations to scale that current L2 networks have are well-known :
>> broadcast storms that can propagate along the switched domain, spanning
>> tree convergence (argument to be reviewed with RSTP), excessive ARP
>> broadcast difussion, inefficient links usage by spanning tree algoritm.
>> These problems increase with L2 domain size and need to be alleviated.
While I agree with these issues, the point is that it is NOT the goal of
rbridges to address all of them.
Rbridges focuses on inefficient link use by spanning trees, and also
addresses convergence time.
It does NOT address broadcast storms either due to regular broadcasts or
address resolution. I'm not sure an L2 solution CAN deal with these; the
former is asking to flood, and flooding doesn't scale. The latter can be
replaced with centralized solutions (e.g., as was done in LANE for ATM),
but at the cost of robustness and zero-configuration.
>> Besides this, the Ethernet domains tend to grow in actual campus
>> networks: more devices (sensors, PDA, displays) connected. My
>> understanding is that Rbridges should try to improve this situation and
>> allow bigger network sizes under a single IP subnet.
The goal of an rbridge is better BW first, and hopefully better
convergence. This MAY have the side-effect of encouraging larger
installations, which should be addressed -- but supporting larger
installations is not a primary goal.
rbridge mailing list
rbridge at postel.org
> rbridge mailing list
> rbridge at postel.org
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 254 bytes
Desc: OpenPGP digital signature
Url : http://www.postel.org/pipermail/rbridge/attachments/20050927/7e4f06f4/signature.bin
More information about the rbridge