Enterprise Technology By Phil Edholm

Wall Street WAN Managers to Eliminate Routers?

I was just in a meeting with the wide area network managers for a large Wall Street investment firm. These are the guys who architect the router network, CCIEs and all. As we talked through the future of their WAN, one of them said something I never thought I would hear as an unsolicited comment: "What we really want is a way to send L2 packets over the WAN as a connection…so we can just have a simple Ethernet Switch in the branch…eliminating the router altogether." I was astonished, after 5 years, the router-less branch is beginning to be seen by network managers, even by the router aficionados, as a potential reality. Back in 2002, the concept of a simple branch with L2 connectivity over an "Optical Ethernet" seemed interesting, but not practical.

However, maybe that is all changing ... most of the traffic from the branch is to the core (data center and real-time), so intelligence and all that jazz is not really required. Add in the advent of PBT as a metro opens up the capability of the carriers to offer simple L2 Wan networking with QoS SLAs.

So the question emerges; is the future of the branch to be a wiring closet at the end of a string? Will the services be centralized and the branch be a L2 Ethernet switch connected to a Ethernet metro/WAN service? If so, will the branch management be a simple as a wiring closet? What do you think ... is their a router-less world in your future? Is this desirable? Is it practicable? If so, when?

Comments

  1. Phil,

    I can tell you that the very large customer I consult with sees switched ethernet as the only manageable solution. MPLS to tough to manage on networks resembling their scale, so the answer is yes nortel was way ahead of the market (by more than five years) with OE.

    I was just as surprised when I got out in the real world and understood that JR’s “internet time” was just another case of nortel marketing internally to itself :)

    Branch management should be *simpler* in a lot of ways than a wiring closet, because it should be (largely) wireless. If the proper security is in place, the answer is it is not only practicable and desirable, but it is happening now. Does it look like nortel envisioned it? Not entirely, but pretty close.

  2. Is is possible? Is it practical? Yes. We have a remote campus 1,600km away connected on an Ethernet vlan with the routing done on our central LAN 8600’s. Removed all the IP addressing issues, VPN performance issues, etc. The remote campus just appears as part of our LAN as it should. Full gigabit speed for imaging, etc.

  3. Carl, you have obviously taken the step that I think many will follow over the next few years as Etherent VPN services become available. It would be interesting to characterize the value this delivers to you. Is the value quantifiable in cost avoidance or capability? If one branch has value, what is you were a bank with 1800.
    Another question would be how this affected your purchase decision of the service offer to connect that site? What economics and value enabled this decision?

  4. Will the services be centralized and the branch be a L2
    Ethernet switch connected to a Ethernet metro/WAN service?
    This would be nice! Most of my 100+ wan sites are just Stub’s. Why wouldn’t we want to eliminate the WAN Router. We are working on laying our own fiber to remote sites but it is very expensive. Fiber is the only option, other than MPLS, which is expensive.

    What do you think … is their a router-less world in your future?
    My environment is ready for this..Full Mesh Square SMLT with 8600’s. But how would you connect via L3 to these remote sites without Fiber? Do you have some kinda of special switch that emulates a sync interface? I know Tasman has a L2 forwarding feature, keeping the router at Layer 2. Couldn’t you just build up a 5500/470 series switch with an csu/dsu interface and configure L2 forwarding?

    On another note:
    We want to eliminate the service provider but Fiber is expensive.

    Is this desirable?
    YES!

    Is it practicable?
    If it is cost effective.

    If so, when?
    Next budget cycle (08/09)!

  5. Dedicated fiber is expensive, but with virtualized L2 services becoming available, this may be a real possibility. The challenge will be whether it is limited to urban areas or extends out farther.

  6. In order to answer the question, we have to examine the how routers are used today, and the features that are commonly invoked to address the ‘problems’ that enterprises encounter when deploying large, mission-critical networks.

    Since a router is nothing more than a ‘box of software tricks’ that address various problems related to scalability, management, security, class of service, policies, etc - we may find that we can address many of these problems at layer 2.

    If we find that we can address the problems that enterprises face using layer 2 devices, and that it is more cost effective to do so, then the industry will take notice.

    Perhaps Nortel will need to create a layer 2 reference architecture for a very large branch network and prove that it will be functional, reliable, secure and cost effective.

  7. That is an excellent idea.

  8. Phil,

    I would agree with a “brand” or reference architecture that is similar to the Cisco “Access, Distribution, Core” model and goes deeper showing how products and technologies can move traffic securely and efficiently between the elements. Building product documentation around these concepts is another important area Cisco excels at. This helps their sales, technical and support people to “sing the same song” to the customer

    On another subject (somewhat related), has nortel explored the area of software radios? Now that costs are coming down getting the software closer to the antenna seems to be more and more feasible. I think there are good possibilities for breakthrough products in this area. It seems to me to be a natural for nortel

  9. Mark,

    When it comes to L2 services it seems to me that there are basically two types of customers; ones that want (or need) to manage their own LAN/WAN and related soft and hard infrastructure, the second is content to allow a service provider manage the WAN (and possibly LAN) soup to nuts.

    Fibre costs can be amortized over time by using the second model (if the service provider sees profit in running the fibre). As well, service providers that are “utilities” can usually deploy fibre at less cost that a private concern securing right of ways and sub-contractors. As well there is a certain “future proofing” when the utility “insures” the path and technology.

    OTOH rolling your own network allows for a quicker reaction to change and the possibility to lease excess capacity to others. This can recover some (if not all) of the original outlay for the fibre.

    Depending on your location, you may not have the luxury of both choices. It seems to me that if you don’t have both options there might very well be others in the same boat and there are opportunities for defraying costs by sharing Layer 1 transport.

  10. I assume if you did directly connect a ethernet WAN service into switch (sans router) you would want a layer 3 capable switch? I am assuming you do not want DHCP and other bcast traffic crossing the WAN?

  11. Dustin,

    DHCP traffic is pretty minimal (especially in the case of normal lease renewal), unless the DHCP server is down. If you have multiple DHCP servers for capacity or redundancy I would not think that controlling DHCP traffic would worth the effort. A Single DHCP server on a subnet that always responds to the first request from a DHCP client generates 1,368 bytes, or 10,944 bits, in four messages. One-fourth of these are broadcast packets. This would only occur on startup and likely subsequent requests are renewals which generate half that traffic every 36 hours or so (none of it broadcast).

    Other broadcast traffic can be contained on different vlans and is restricted to individual segments, preventing collisions between domains. Broadcast storm protection can also be implemented effectively through vlan segmentation.

    Of course there are ways to do these things at L3 (e.g. spanning tree) as well.

Leave a Reply