[Hipsec] AC/ACR or NES based RR?

[Spencer Dawkins" <[e-mail] (Spencer Dawkins)]

See the multi6 archives, especially the thread from October/November
on "Re: Notes on draft-crocker-mast-analysis-01.txt", and anything
posted by Mark Allman.

Spencer

From: "Jari Arkko" <[e-mail]>

>
> > Considering a case like using GPRS (900 ms delay,
> > real throughput of maybe 10 kilobits/second) and
> > WLAN (20 ms delay, 5 Mb/s), it is clear that
> > different SAs are needed in both directions.
> > Otherwise any packet sent over the GPRS channel
> > will get dropped at the receiving end as outside
> > of the replay protection window.
>
> True. But there's also a corresponding transport
> layer issue, even if you get the SAs right. If you
> were carrying a single TCP session, the disparity
> in the interface speeds would cause problems for TCP.
> Hence it may make more sense to constrain the use
> of multiple interfaces for failover rather than
> load balancing, or even randomly using more than
> one interface. Has this been discussed previously?
> Are we talking about this yet anywhere in the
> documents?