[c-nsp] MTU / BGP

[Mark Tinka]

On 21/Jul/15 10:42, James Bensley wrote:
> This is more or less what I was going to say, there is an underlying
> MTU problem in your network. Although an MTU setting per neighbour
> knob would be nice, its only really needed as a bit of a fudge. Its
> more ideal to have consistent MTUs everywhere.

We have found keeping TCP MSS to 1,500 bytes to work well across the board.

It's a bit of a sacrifice in the core backbone where we run a 9,192 byte
core, but at 1,500 bytes, we still get decent BGP update message
transfer for today's full IPv4/IPv6 table.

Not having to deal with custom per-node MTU settings to take care of BGP
is a big plus, operationally.

Mark.