Segment Routing Anyone?

One of my readers listened to a podcast where a $vendor described how they found another use case for source routing IPv6 segment routing (SR): 5G networks… and wondered whether SR made a comeback or is about to.

I don’t know nearly enough about mobile networks to have an opinion, however…

Considering that mobile network backhauls use tunnels of some sort anyway, using source routing (which is what segment routing really is) instead of tunnels might make sense, although it still sounds pretty academic to me.

Quick Diversion Into SR 101…

Keep in mind the networking vendors managed to squeeze two completely different technologies under the Segment Routing umbrella:

  • MPLS labels stacks,
  • IPv6 extension headers.

You can use segment routing with MPLS label stacks on existing hardware (the only limit is the stack size on ingress node), while IPv6 SR needs either new hardware or NPUs.

At least the MPLS version of Segment Routing uses new control-plane extensions to existing routing protocols (IS-IS, OSPF, BGP) to distribute labels for prefixes (including nodes when prefix = loopback interface), or adjacencies (read: interfaces unless you’re using shared LANs).

And We’re Back…

MPLS version of segment routing seems to be popular in large data centers, particularly at WAN edge of large content providers. It’s also a relatively small evolutionary step - data plane is the same, control plane is replaced with something that makes way more sense than current hodgepodge of LDP, RSVP…

Even outside of that particular use case, using segment routing functionality to allocate MPLS labels to BGP next hops makes perfect sense and eventually simplifies your network.

IPv6 Segment Routing is another story. It replaced the abhorred (and now deprecated… because nothing ever dies in IETF) RH0 with something else, but it’s still source routing using extension headers… and very few hardware platforms support it (all of them with special engineering code).

The whole IPv6 SR saga seems like another LISP exercise to me - we have a technology we invested a lot of money in, let’s find a problem it could solve. Look, here’s another one…

Am I wrong? Is someone using IPv6 SR in (anywhere close to) production?

12 comments:

  1. To continue the conversation, Packet Pushers have a (Sponsored) Podcast about Segment Routing v6 : https://packetpushers.net/podcast/show-370-cisco-segment-routing-ipv6-sponsored/
    (No Strings Attached)
    Replies
    1. Where do you think the whole idea originated from? There's only one place sharing such ideas ;))

      With all due respect to all the good work Packet Pushers are doing in their many podcasts, sponsored content = $vendor telling us what we should buy, regardless of how well it's sugarcoated in a technical discussion... and more so if the $vendor happens to be the only one selling said technology. That's why I stopped doing sponsored webinars.

      As for "no strings attached" to sponsored content, ask yourself who is the customer and what (or who) is the product.
    2. I took part in the NXTWORK 2018 and this idea was also promoted by Juniper. Probably for the same reason.
    3. Were they promoting SR-over-MPLS (which most everyone is promoting - Arista, Cisco, probably also Juniper), or SR-over-IPv6, which in March (the latest interoperability report I found) was implemented only on some Cisco boxes running special engineering code.
    4. Just looked into the slides presented during the mentioned meeting: the segment routing presentation mentioned at the beginning that "segment encoded as the label or IPv6 address" but frankly speaking I do not remember any details.
    5. If you listen to Packet Pushers podcast ... you can hear directly from Bell Canada and LinkedIn on why they think SRv6 matters to the industry.
  2. IPv6 Segment Routing is supported by recent Linux kernels, which makes it possible to deploy it inside enterprise networks and datacenters. See https://www.segment-routing.org for some use cases that have been made possible thanks to this implementation, including network programming.
    Replies
    1. Linux has also supported MPLS and VXLAN for years (and VXLAN is offloaded by all server NICs while other encaps aren't). So I have to share Ivan's skepticism. Your scientists were so preoccupied with whether they could, they didn't stop to think if they should.
  3. Seems like history repeats itself. What about the security implications of source routing with IPv6 extension header? Is IPv6 / MPLS SR forwarded in hardware or software? Does it need a custom ASIC? I found use cases: https://tools.ietf.org/html/rfc8354 There's a lot you can do with SR. It looks like a new business model.
  4. There is hardware support for both MPLS and v6 at this point. Full disclosure I now work for Cisco, but the amount of people looking at SR is a very long list these days. There is a reason why every major vendor now supports SR-MPLS. SRv6 has a longer road but Juniper recently made a point to mention it in their Penta chipset announcement.
  5. IMO, all main vendors will support SRv6 in HW sooner or later just because... The SRv6 saga will be also dominant in everyone's talks. Besides that support of SRH and IPv6 are needed for that (technical side), I see (I also work for another vendor :) but this is just my personal opinion) the main benefit of SRv6 is in potential distributed network programmability (function+locator encoding in SID) as it is described in draft-filsfils-spring-srv6-network-programming.

    I have not seen SRv6 in production cases yet in my region, but people talk about it.
    For average operator which has traditional MPLS based network, IMO, main concern is a justification of all SRv6 migration spendings and efforts. For mobile SPs, 5G could be such potential case. But again - as Ivan often says: "it depends" :)
  6. I haven't seen any sign whatsoever that Juniper supports SRv6 (& I was at NXTwork :-). They're definitely on board with SR MPLS (in IS-IS, OSPF, & BGP) & have been for a while.
Add comment
Sidebar