Dear RTGWG
I was wondering if folks here have some knowledge/data of our current
understanding,
of the behavior/performance of strict hop-by-hop source routing.
I am asking, because in application where we need strict path guarantees like
latency or
throughput, such as in DetNet, we do operate on the expectation that we can
nail down the
path completetely (hop-by-hop), and any type of path failure would also have
pre-determined
reactions (none when we use path diversity with PREOF, or pre-established
tunnels, when
we use RSVP-TE, but of course, there are also SR variants of that).
In any case, to scale, strict hop-by-hop source routing was AFAIK for decades
not an attractive
option because of the header becoming too large, but with compressed Segment
Routing headers
"around the corner" ( ietf timeline of SPRING ;-), i thought we would want to
reinvestigate
that option as an industry.
Asking some old-timers, my impression was that we should be able to provde
competitive solutions
with this approach that would not have intrinsic isues of hop-by-hop routing,
such as microloops,
and that we should also be able to improve convergence speed quite a bit,
because most of the
routing protocols performance limits come from download routes to hardware
forwarding on every
hop. If the IGP instead does not have to do this, but only needs to flood
routing information
at the control plane, it could likely be speed up quite a bit. And the
installation of
source routie header changes resulting from IGP routing changes would likely go
mostly to
software based PE in many networks, or else only has a likely much smaller set
of entries
required.
So... Was wondering about any thoughts about this by group members
Cheers
Toerless
P.S.: Oh, and yes, i also have the interest from the perspective of also
working on multicast
hop-by-hop steering solution ;-))
_______________________________________________
rtgwg mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/rtgwg
