Is possible RSTP convergence within 50 ms on a ring topology?

Unanswered Question
Aug 5th, 2007

I've been trying to lower the price point for a customer by using a new topology on their network.

They are intending to use MPLS fast reroute just because they have been told the convergence time goes down to 50ms.

Of course, the advisor would gladly provide them with the required hardware upgrade from 3560 to 65xx switches.

I am tellim the customer that as long as they keep the LAN Ethernet 1000SX based ring topology with two exit routers and all other ports marked as edge ports, there is no need for MPLS because:

- topology changes such as link down and switch down are propagated inmediately with no waiting time, all delays involved are protocol procesing on the switch and negotiation between switches

- RSTP may no be as fast as 50ms in their scenario but its convergence time would be in the range of 100-200ms at a 10% the cost of the MPLS FRR option (far beyond their needs).

Am I right? Can you think of any way to find out the true convergence time without disrupting their network? The ring is made up by two 7600 as exit routers connected to the RSTP ring on two of thirteen Catalyst 3560 on the ring.

I have this problem too.
0 votes
  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 5 (1 ratings)
drolemc Fri, 08/10/2007 - 06:51

I think it is not possible RSTP convergence within 50 ms. But ring wrapping on the ML-Series card allows convergence times of less than 50 ms for unicast and pass-through traffic. Cisco proprietary RPR convergence times are comparable to SONET/SDH and much faster than STP or RSTP. Also I think use Spanning Tree; it would enable loop-free topology. A logical tree topology is created on the physical ring with shortest paths to the Root. Use 802.1w RSTP is good choice.

jmartinez3 Fri, 08/10/2007 - 08:17

Hi, drolemc.

I already had a look at it but RPR works on ONS devices only, much more expensive than RSTP switches. That over-investment is what I am trying to avoid.

Further research has shown that BPDU prcessing in CPU is soooo slow that RSTP cannot converge any faster than 50ms per hop.

Looks like other vendors are developing cheap BPDU hardware engines so convergence is no longer limited by switch processing power, but protocol delays and waiting times. That lets RSTP compete against SDH, RPR and MPLS FRR at the price point I was looking for.

Hope that comes out soon.

By now, I could not find a viable option.

Thanks a lot for the interest.

wochanda Fri, 08/10/2007 - 08:39

We've got something in the works that will make this happen for you. Keep your eyes open for the 12.2(40)SE release for 3560/3750.

jmartinez3 Tue, 08/14/2007 - 06:51

Hi, William.

I could not find the projected release date for IOS 12.2(40)SE, could you please provide me with an aproximate date or where to find it?

Thanks in advance (and looking forward that IOS image). :)

jmartinez3 Tue, 08/14/2007 - 07:29

I googled the IOS name instead of searching inside and I found the answer. IOS release 12.2(40)SE is due 24 August 2007.

Sorry for asking before googleing it.

jmartinez3 Thu, 08/16/2007 - 07:03

I guess you are not more specific because you cannot be so it would be useless to ask about a bit more information on that, am I right?


This Discussion