Juniper's Packet-Optical Spells M-P-L-S

Juniper Networks Inc. (NYSE: JNPR) has a plan for the future blending of packet and optical networks, and it doesn't necessarily jibe with what others, such as Alcatel-Lucent (NYSE: ALU), have in mind.

Juniper, which paired up with Nokia Networks for its packet-optical plans, envisions a core network where Multiprotocol Label Switching (MPLS) label-switched paths (LSPs) aren't determined on the fly by routers, but are provisioned through the network core just as wavelengths are today.

"You make the MPLS layer the transport layer. That's where I think the packet-optical discussions are going," Juniper senior director Luc Ceuppens tells Light Reading.

And he's not talking about MPLS-TP. In Juniper's thinking, MPLS-TP comes into play in an all-OTN network, when large ODU payloads are being shunted around. Juniper is talking about more finely grained units of traffic, carried over the optical transport network via MPLS.

It's not a popular view. Many companies see packet-optical transport systems (P-OTS) and converged packet-optical networks as a way to use fewer router ports in the network core. Verizon Communications Inc. (NYSE: VZ) went so far as to ask the industry to create new long-haul optical boxes that could let traffic sometimes bypass router ports, and Alcatel-Lucent has crafted a packet-optical strategy along those lines. (See Verizon Rethinks Long Haul and AlcaLu Makes Its Packet-Optical Move .)

These companies argue that a majority of the traffic going through a router doesn't need to be processed, because its destination is farther along in the network. So, why not pack that "express" traffic into wavelengths that skip the router?

Guess what: Juniper doesn't see so much of a need to bypass routers. "We believe a lot of the bypass cases are corner cases," Ceuppens says.

He says router bypass might work at an Internet exchange point, which deals in huge chunks of aggregated traffic. But he argues it's less efficient in most cases, "when you have tens of thousands of LSPs on the network" all headed to different destinations. Router bypass, in that case, would "end up adding wavelengths that are then not filled in the most efficient way." The result: What you save in router ports would be spent adding more P-OTS ports.

(Juniper claims to have a study that verifies this; look for the results to come soon, Ceuppens says.)

To Juniper, the real benefit of packet-optical convergence is that MPLS, once married to the optical network, could be made more deterministic. The network could provision MPLS connections the way wavelengths are provisioned.

"Very often, when we talk to our customers, they bring this up," Ceuppens says. "They like the MPLS network, but they want the manageability they have in their TDM networks."

Whether this idea takes off might depend on how many carriers Juniper can convince. Verizon is dead-set certain that it can save money by using P-OTS to bypass routers some of the time, and much of the industry seems to back that notion, according to Heavy Reading analyst Sterling Perrin.

"Operators tell us 60 percent of router traffic is transit," says Perrin. "There's a belief that stuff going through the router doesn't need to go through the router -- and that routers are more expensive than transport equipment."

Another factor to consider is that MPLS-TP, which would provide some of the determinism Ceuppens is talking about, appears to be gaining popularity as Provider Backbone Bridging - Traffic Engineering (PBB-TE) loses ground. "We are seeing things jelling around MPLS-TP right now for carrier-oriented Ethernet," Perrin says. (See MPLS-TP vs. PBB-TE.)

Ceuppens says the argument between the two comes down to where a carrier wants to do its switching: in routers (Juniper's MPLS case) or in optical transport (the OTN case that more of the industry seems to be pursuing).

They aren't mutually exclusive cases, and Ceuppens sees reason for the industry to develop a box that does both, switching either MPLS or OTN traffic natively as it arrives. That would be a P-OTS, of course -- although Ceuppens thinks such a box should be defined by having a single switch fabric that handles both traffic types natively.

Important point: Juniper and NSN have announced no such system, nor have they even said they intend to develop a P-OTS: Ceuppens declined to comment on any P-OTS plans. It's worth noting, though, that Juniper has invested in P-OTS startup Cyan Optics Inc. (See Cyan Plays God With Optical.)

— Craig Matsumoto, West Coast Editor, Light Reading

Interested in learning more on this topic? Then come to Ethernet Expo 2009, Light Reading’s ninth conference and exposition covering the hot topic of Carrier Ethernet network technologies and services in North America. To be staged in New York, November 3 & 4, admission is free for attendees meeting our prequalification criteria. For more information, or to register, click here.

digits 12/5/2012 | 3:53:46 PM
re: Juniper's Packet-Optical Spells M-P-L-S

Does Juniper have its head stuck in the sand, or is it the one company really understanding how packet/optical architectures are going to play out?

tmmarvel 12/5/2012 | 3:53:45 PM
re: Juniper's Packet-Optical Spells M-P-L-S

While I don't totally agree with Juniper's router-centric network view, I think the JNPR guy is right on the following point:


q' Router bypass, in that case, would "end up adding wavelengths that are then not filled in the most efficient way." The result: What you save in router ports would be spent adding more P-OTS ports. 'q


However the flip side of this is that whatever router/switch-heavy network saves on wavelengths and optical equipment ports (by filling them more efficiently through packet stat muxing at each network junction), is spent on the router gear that gets more expensive the more finer the level of traffic processing.

So what gives - is the architecture work on finding best mix of routing, switching, muxing & optical transport sort of in vain if the efficiencies of any given architecture at one level are negated at another level?

Are the best savings then just in minimizing any spend on architecture work, or is there room for real efficiency gain and how?

dcdragos 12/5/2012 | 3:53:45 PM
re: Juniper's Packet-Optical Spells M-P-L-S It appears Juniper wants to maintain their relevance even in the optical-dominated P-OTS/OTN world where they do not really play with their routers. So, to avoid a router bypass, it seems they are reinventing PBB-TE, using MPLS labels instead of Destination MAC+VLAN combinations (as in PBB-TE), and maintaining the transport paradigm of statically provisioning MPLS "cross-connects" for a certain MPLS label on each node on the path. Unless other big router players endorse this, it will be restricted to a specific play in the nework.
Huub_van_Helvoort 12/5/2012 | 3:53:42 PM
re: Juniper's Packet-Optical Spells M-P-L-S

The statement "In Juniper's thinking, MPLS-TP comes into play in an all-OTN network, when large ODU payloads are being shunted around. Juniper is talking about more finely grained units of traffic, carried over the optical transport network via MPLS"shows that Juniper does not understand the concept of MPLS-TP: there is NO relation between MPLS-TP and large ODU payloads or any other lower layer technology.

The intention of MPLS-TP is to provide the same OAM behaviour as other transport technologies (e.g. carrier class Ehernet, SDH, OTN) such that the service provider staff can maintain the network accross all technologies in the same way. 

mvissers 12/5/2012 | 3:53:42 PM
re: Juniper's Packet-Optical Spells M-P-L-S

There are indications that it is beneficial to replace a Label Switched Path (LSP) by a sub-Lambda Switched Path (sLSP) as soon as that LSP has a bandwidth of 0.4Gbit/s or more. G.709 (OTN NNI) has been extended in 2009 to provide rigth sized sLSPs by means of a new flexible ODUk, referred to as the ODUflex. Such ODUflex can have any bit rate in the range from 1.25 to 100 Gbit/s. When carrying packets, bit rates of n x 1.25Gbit/s (n=1..80) are recommended. Multiple sLSPs are carried over one (10G, 40G, 100G) wavelength ("Lambda Switched Path"), filling this wavelength efficiently.

With the growing amount of video content, some operators already expect to have most of their core domain LSPs exceeding this 0.4Gbit/s threshold in the next two years. Those core networks will have replaced their LSPs in 2011 by sLSPs.

sLSPs start/end on the L2 and L3 PE nodes and are transported through the emerging multi-service, switched optical transport network (OTN).

sLSPs carry the L2 and L3 service layer packets (i.e. Ethernet and IP) directly via Generic Framing Procedure (GFP) based encapsulation. Lambda Switched Paths carry the sLSPs via Generic Mapping Procedure (GMP) based encapsulation.

The packet optical signals in those core networks will then contain the following protocols: Ethernet or IP, GFP, sLSP, GMP and LambdaSP.

Huub_van_Helvoort 12/5/2012 | 3:53:41 PM
re: Juniper's Packet-Optical Spells M-P-L-S

IF (and this is a big IF) they are reinventing PBB-TE then I don't understand their fierce opposition to reuse the OAM described in IEEE 802.1ag and ITU-T Y.1731.

Pete Baldwin 12/5/2012 | 3:53:41 PM
re: Juniper's Packet-Optical Spells M-P-L-S

Good discussion, folks.  I like Dcdragos's comparison to PBB-TE.

"With the Lights Out," I think you're asking the right questions.  Luc Ceuppens was saying that part of Juniper's argument rests on the fact that router ports will get cheaper as time goes on ... apparently, if you do the math while taking that into account, you get an answer closer to Juniper's.  I'll be interested to see this study they're working on.

jayjay 12/5/2012 | 3:53:39 PM
re: Juniper's Packet-Optical Spells M-P-L-S Why reinventing PBB-TE. Statically provisioned connections are a general transport network feature and can also be done with MPLS-TP. So no need to reinvent PBB-TE. MPLS-TP and Optical/ODU switching do not compete as such they complement each other. It is only the question when you switch over from packet to ODU/wavelength switching. In this a lot of factors play a role like equipment costs, network topology, traffic patterns which will even change over time.
tmmarvel 12/5/2012 | 3:53:38 PM
re: Juniper's Packet-Optical Spells M-P-L-S

Regarding the prospects of JNPR's (or other's) router ports getting cheaper over time (presumably vs plain digital/optical mux ports per b/s), it would be important to see what underlying factors would be driving toward that direction.

Intuition tells that it would be the other way, that routing and packet switching cards doing L3-L1 processing will get more costly (including to operate, consider management o/h, power consumption etc) relative to plain L1 or L0 transport cards per interface bit rate.

But as noted, the processing cost per transmitted bit rate is not the relevant factor to consider, instead it is the cost per revenue generating bit (assuming all actual payload bits carry same revenue level), and thus it also matters how efficiently or inefficiently the interfaces, cheap or not, get filled with actual payload traffic.

If there is an answer to better overall efficiencies, it appears that it would involve finding a way to improve the network usage efficiencies while keeping the carrier network processing level to minimum, ideally just L1 or L0, besides the provider/system domain boundaries where routing or gateways are needed.

That may not be what router-centric vendors wants to promote, at least until one would have an advantage in the new router/gw-light provider core networking; perhaps looking at this new direction might give JNPR advantage, considering eg



Sign In