MPLS Vendors Demo Fast Reroute

Yesterday, Isocore, an independent testing facility in McLean, Va , hosted its third public interoperability demonstration where it tested for the first time a new Internet Engineering Task Force (IETF) draft of MPLS Fast Reroute, a technology that brings Sonet-like protection to Multiprotocol Label Switching (MPLS) networks (see MPLS Fast Reroute Gains Momentum).

Cisco Systems Inc. (Nasdaq: CSCO), Avici Systems Inc. (Nasdaq: AVCI; Frankfurt: BVC7) and Juniper Networks Inc. (Nasdaq: JNPR) participated in the test using specifications from a draft written by George Swallow of Cisco, which combine work from two other drafts on Fast Reroute. The test focuses on link protection and did not include node protection mechanisms. Cisco and Avici had tested an earlier Fast Reroute draft for link protection in August.

Ixia (Nasdaq: XXIA), Spirent plc (NYSE: SPM; London: SPT), and NetTest provided the testing equipment.

The importance of the demo is twofold. For one, this is the first time that all three core routing vendors have tested this feature together. And second, it is an important step forward in developing this critical feature for MPLS.

“The idea behind doing these demos is to show certain technologies working today,” says Bijan Jabbari, president of Isocore. “But we need to work further to make sure all the features are standardized. And I think that this process will help facilitate that.” The test bed included IP core routers from three vendors: Avici’s TSR, two Cisco GSR 12404s, and Juniper’s M10. The way it was set up was that traffic entered the network at the CoSine Communications Inc. (Nasdaq: COSN) IPSX3500, which was used as the access router. It routed traffic to the Cisco GSR. The Cisco GSR then routed it to the Juniper M10, which routed it onto another Cisco GSR. When the cable was disconnected between the Cisco GSR and the Juniper M10 router, the Fast Reroute mechanism kicked in and rerouted the traffic to the Avici TSR, which was connected to the GSR and M10. Is that clear? The network restoration occurred in roughly 5 milliseconds, well within the current Sonet standard that requires 50 milliseconds restoration. Restoration in the demo has consistently run within 10 milliseconds, says Jabbari.

The reason the restoration is so quick is because the traffic is rerouted locally. There is no signaling protocol that propagates through the network to let the next hop know packets are coming, making the failover time to an alternate path extremely fast, says Don Troshynski, senior systems engineer for Avici.

The demonstration itself was not terribly exciting to watch. In fact, most of the observers didn’t even realize the test was going on until it was over. Even Tony De La Rosa, product manager for Ixia, who was running the demonstration and explaining what was happening, hadn’t realized that the line between the routers had been disconnected. Interestingly enough, this seems to be the point of demonstrations showing off reconvergence technologies. The point is to show restoration occurring so quickly that there is no noticeable loss of service.

“Restoration time can fluctuate within a certain range,” says Amrit Hanspal, product manager for MPLS and QOS at Cisco. “As long as you keep it below 50 milliseconds, it’s alright. And even then some applications will be fine if the restoration time is longer. Call signaling is usually dropped after 1 or 2 seconds.”

The vendors involved in setting up the network and the test have been working together on this project for the past few months, dedicating two full days to setting up the demonstration prior to its public debut. The work that was done to get this implemented will go a long way in helping these vendors further define the standard, says Loa Andersson, chief architect for Utfors AB.

“From an IETF perspective this is a pretty good basis for developing the protocol design,” he says. “It’s still early days for this feature, and this a good way to work out the problems early on.”

Andersson says that from a service provider perspective, this demonstration could highlight problems that he would look for when implementing the feature in his company’s own test bed.

Heidi Tse, a backbone engineer for WorldCom Inc. (OTC: WCOEQ), says that these demonstrations are important to a degree, but she emphasizes that carriers must still test the gear themselves.

“We don’t want to be a vendor’s guinea pig,” she says. “But you also have to understand that this is not a sophisticated test, and we would have to test it much more in depth ourselves if we were going to implement this feature. Still, tests like this are important because it will help move the standard along and we all need the standard before we do anything.”

Isocore also demonstrated yesterday interoperability for label switched path (LSP) protection using gear from Marconi plc (Nasdaq/London: MONI). And it demonstrated provisioning of Layer 3 MPLS/BGP (Border Gateway Protocol) and Layer 2 Draft Martini virtual private networks using gear from Cisco, CoSine Communications Inc. (Nasdaq: COSN), Foundry Networks Inc. (Nasdaq: FDRY), Extreme Networks Inc. (Nasdaq: EXTR), Lucent Technologies Inc. (NYSE: LU), and Redback Networks Inc. (Nasdaq: RBAK).

Jabbari says Isocore plans to conduct more interoperability tests next year. Specifically, he would like to test node detour in MPLS fast reroute in addition to the link protection that was demonstrated already. But he says he must wait for all the vendors to have a working implementation of that spec before a test can be designed. Andersson and Tse will likely be tuning in again to that one.

“Node protection is on my Christmas wish list,” says Andersson. “But we need to take one thing at a time. I’m glad they’ve got the link protection working. Now they can move on to the node protection.”

— Marguerite Reardon, Senior Editor, Light Reading
Page 1 / 16   >   >>
achorale 12/4/2012 | 9:26:07 PM
re: MPLS Vendors Demo Fast Reroute This is the maliase of the days - anytime some folks demonstrate some good results, people strip this apart!

I guess ignoring propagation delays, this was a great result. Add in whatever propagation delays you want and once can see that the resultant failover time is impressive.
lob 12/4/2012 | 9:26:07 PM
re: MPLS Vendors Demo Fast Reroute Now we have absolutely meaningless results (5-10 ms) achieved in a test network with ridiculously simple topology and with no speed-of-light propagation delays.

How illuminating. Bet you could do just as well with plain old I-ISIS or OSPF. Just don't use link keepalives as the sole means of detecting LOS.

Running Dijkstra algorithm for 6 nodes on anything more advanced than Z80 would likely take less than 5 ms :)
Tony Li 12/4/2012 | 9:26:06 PM
re: MPLS Vendors Demo Fast Reroute In case you missed the point, the speed of light propagation delays are completely irrelevant in fast reroute because there is NO signalling from the hop before the failure back to the head end of the LSP.

Yes, you could do as well with an IGP _in this topology_, but this result holds independent of the specific topology. The failover time here is simply the amount of processing necessary to install the backup LSP. This time should scale linearly in the number of LSPs that have failed.

arch_dude 12/4/2012 | 9:26:06 PM
re: MPLS Vendors Demo Fast Reroute lob says:

Now we have absolutely meaningless results (5-10 ms) achieved in a test network with ridiculously simple topology and with no speed-of-light propagation delays.

How illuminating. Bet you could do just as well with plain old I-ISIS or OSPF. Just don't use link keepalives as the sole means of detecting LOS.

Running Dijkstra algorithm for 6 nodes on anything more advanced than Z80 would likely take less than 5 ms :)
lob completly misses the point. MPLS fast reroute reacts locally. Therefore its reaction time is independent of the network size and of the network topology. In this regard it is just like SONET. By contrast, IGP convergence times depend on the size and topology of the AS.

What this means is that the fast reroute results in a big network will be the same as the fast reroute results on the test network.
beowulf888 12/4/2012 | 9:26:05 PM
re: MPLS Vendors Demo Fast Reroute lob:
I don't think speed-of-light propagation delays would affect the FRR time. Correct me if I'm wrong, but FRR doesn't provide end-to-end protection for an LSP. How is this any different from SONET APS? FRR and APS both rely on the detection of hardware failures.

Now if this were a test of backup LSPs, I'd have to agree with you. Of course, the details of the testing regimen weren't revealed in the article...

best regards,
techstud 12/4/2012 | 9:26:04 PM
re: MPLS Vendors Demo Fast Reroute
MPLS by now is an obsolete junk - who cares, except for certain vendors and individuals who have vested interest to sell their gears or so called expertise (if re-writing old atm papers can be called any thing). The reasons mpls work was started in IETF are no longer applicable.

Only useful tauted application for mpls is carrier bgp/vpns which is the most convoluted solution to a problem for which many other elegant approaches exist. Who needs traffic engineering at 10% utilization of network capacity (in any case there are simple approaches to route around congested links)? DWDM gear is sitting unused if you need more bandwidth and 80% (??) fiber is sitting unlit. So why care these tons of mpls junk?

Whole tag switching at cisco was a reactionary development to deal with the threat from Ispilon's fast packet switching etc. It wasn't based on any revolutionary research. Route look up technology has since that time has matured a lot since 1997 when all this started(software algorithm and hardware loop up engiens - NPS, CAM and specailized implementation). Most mpls drafts including all re-route, crank bank schemes, traffic engineering etc are rehash 80/90s atm/circuit switching papers on similar topics.

Mpls is dead, backward looking technology except that now it comes in new shining packaging and with bold letters "NEW". users beware.

sgan201 12/4/2012 | 9:26:03 PM
re: MPLS Vendors Demo Fast Reroute Hi Tony Li and others,
Speed of light propagation and other stuff do affect in a certain degree. It depends on how the cable is cut. There are two optical cables in each port: TX and RX. The proper test is to only cut the RX cable. The RX side will detect the failure. It will generate Sonet AIS to the TX side on the same port. Depending on the length of the TX cable, the other side may take a while to receive AIS. Obviously, if the TX cable is a few thousand miles long, it will take a while to switch over. How does MPLS FRR ensure that the port is switchover on both side at about the same time???

However, we cannot tell from the article whether the test is done properly by only cutting one cable from the port. Cutting both cables on the same port is a trivial test since both side would have detected the LOS and can switchover immediately...
sgan201 12/4/2012 | 9:26:02 PM
re: MPLS Vendors Demo Fast Reroute Hi,
MPLS is very useful of keeping your competitor investing on useless stuff that no one will buy. Then, you can focused on creating something marketable with little or no competition..
Look at situation right now..
RBOC want to buy ATM switch.
There are only 2 next gen ATM switch start up: Wavesmith and Equipe..
Meanwhile everyone waste their money and effort on MPLS...

When are we going to learn that consensus is always wrong in data comm market??

Remember, 15 to 20 years ago everyone is convinced that computer should be inter-connected via PBX. NT, AT&T, Rolm, Siemen and everyone else dump 100 of millions dollars into that and then Ethernet show up and take over the LAN.

Mech4 12/4/2012 | 9:26:02 PM
re: MPLS Vendors Demo Fast Reroute Great test, great results - who cares.....

The 2 things I get from this article is that these 3 cos. did something together, that worked and that Worldcom still gets their thoughts into public forums.
opticalwatcher 12/4/2012 | 9:26:01 PM
re: MPLS Vendors Demo Fast Reroute "When are we going to learn that consensus is always wrong in data comm market??"

To a certain degree who is right or wrong or whether or not MPLS is "dead-end" is irrelevant.

What matters is what customers with money are asking for. And most are asking for MPLS in one form or another, either short term or long term. So the vendors are going to put it into their equipment. Even Wavesmith and Equipe have an MPLS story.

Page 1 / 16   >   >>
Sign In