[j-nsp] Spine & leaf

[Mark Tinka]

On 25/Jun/18 23:04, Payam Chychi wrote:

> Not sure if I agree with this, this (ospf) certainly would not scale in my
> network. the point being, different use cases, different environments.
> Always design your network to allow for forward progression else you will
> be wasting more time and dealing with more problems

Indeed - there are valid concerns about the scalability of an IGP in
2018. But none of them point to CPU or memory, IMHO. I think the bigger
concern is the amount of FIB needed to maintain a growing IGP network,
as that FIB also needs to carry other information, e.g., MPLS, BGP,
ACL's, e.t.c.

For such situations, RFC 3107 and BGP-LS are worth considering,
especially if down the line, FIB is going to an issue, e.g., think of
several Metro-E rings running full IP/MPLS for control and forwarding,
on thousands of Cisco ASR920's. Each of those has a FIB space of 20,000
entries (all-in), so you can quickly do the math.

Otherwise, if you have 1000's of MX480's running current-generation
MPC's, this isn't going to be an issue... nor do I think it would be an
issue if you had 1000's of servers running your favorite flavour of
UNIX/Linux on decent hardware.

Mark.