[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: successful termination?



Hi Brian,

a couple of comments.

Brian E Carpenter escribió:
Jari,

If I look at both Geoff's and Olivier's (relayed) comments,
it's clear to me that there is work to be done.

I agree that the address-pair selection issue is broader
than just shim6 (and is related to other things, such as
egress router selection).
note that shim6 need to do both ULID pair selection and locator pair
selection (after a failure) and maybe the selection is different for
these cases (the typical example is that when used in conjunction with
MIPv6, probably the CoA is better as a locator  while the HoA is better
as a ULID. OTOH, maybe in general the option is not so different.

In the scope of 6man, they select an address pair that will play both
roles, so the criteria for selection maybe potentially different.

 But there may still be some work
do here, on 'SHIM6 Requirements for Address Pair Selection',
even if the actual specification belongs in 6MAN.

I agree that there is commonality of interest with multipath
transport, but it's far from clear to me that there is
actual overlap in the mechanisms; SHIM6 is a layer 3 shim,
and multipath transport sure seems like a layer 4 shim,
which was a direction that MULTI6 explicitly choose not to
follow up.
I think the story here is something like this.
Suppose we want to do multipath TCP.
One option is to do it a la SCTP, where the MPTCP exchanges all the
alternative addresses and includes different source dest address pairs
to select the different paths.
Now, the point that can be made is that shim6 already handles the
locator pair management which is far from trivial, espcially when
considering the security issues. So, why not let MPTCP to leverage on
shim6 do do the locator pair management and let the MPTCP to actually
decide which segments to send through each locator pair?

I mean, we can use a comibation of the shim6 API plus the forking
capabiltiies and maybe some new extension so that the MPTCP is not aware
of the actual address pairs, but knows that the underlying shim6 layer
has path diversity. So MPTCP simply will add some tags to state that
some segments should fly over one path and some other segments though an
alternative path.
So, in this case, each layer does what i does best. Shim6 handles the
multiple locator pairs, while MPTCP distribute the segments through the
different pahts based on congestion.

Makes sense?

If so, then we may need to do some work so that both mechanisms interact
reasonably.


That said, it's clear that the union of Geoff's and Olivier's
messages add up to a program of work, basically to find out if
SHIM6 is deployable and useful. I think it would be a shame
to pause before doing that.

agree

Regards, marcelo


  Brian