Hi Spencer,
Thanks a lot for the quick reply. Please see below.
-----Original Message-----
From: Spencer Dawkins [mailto:spencer(_at_)mcsr-labs(_dot_)org]
Sent: Thursday, February 08, 2007 12:52 AM
To: ASH, GERALD R (JERRY), ATTLABS
Cc: ietf(_at_)ietf(_dot_)org; General Area Review Team;
avt-chairs(_at_)tools(_dot_)ietf(_dot_)org;
pwe3-chairs(_at_)tools(_dot_)ietf(_dot_)org;
Andy(_dot_)Malis(_at_)tellabs(_dot_)com; HAND, JAMES, ATTLABS; Mark Townsley;
ext Cullen Jennings; GOODE, B (BUR), ATTLABS; raymond.zhang;
lars-erik(_at_)lejonsson(_dot_)com
Subject: Re: Gen-ART Last Call Review of
draft-ietf-avt-hc-over-mpls-protocol-07
Hi, Jerry,
This is easier than it should be... slicing down through the stuff we
already worked out (if I deleted it, I agree with your plan)...
option is the same for both IPCP and IPV6CP. This configuration
option MUST be included for ECRTP, CRTP and IPHC PW
types and MUST NOT be included for ROHC PW types.
Spencer: Is it obvious what the decompressor does if it sees this
configuration option for ROHC PW types? It may be - I'm just
asking. I'd
have the same question elsewhere (in 5.2.2, for example), but
will only ask it here.
Yes. The corresponding text for the ROHC configuration option is
specified in Section 5.2.5. In other sections we specify that the
configuration options are only applicable to specific header
compression formats, e.g., as in Section 5.2.2 for cRTP.
Spencer: there was this theory about testing TCP with "kamikaze" or
"Christmas tree" packets (you set all the options to "1",
whether that makes
sense or not, and see what the other guy does). I think I'm
asking "what
SHOULD happen if the decompressor sees a packet like this".
I'm wondering if we still worry about things like this...
You make a very good point, error legs of course are critical for
whatever erroneous configuration or coding may occur. We can specify
something like this in Section 5.2.1 (Configuration Option Format)
"... This configuration
option MUST be included for ECRTP, CRTP and IPHC PW types and MUST
NOT be included for ROHC PW types. A decompressor MUST reject this
option (if misconfigured) for ROHC PW types and send an explicit
error message to the compressor [RFC3544]."
and loss, it is still the protocol of choice in many
cases. In these
circumstances, it must be implemented and deployed with
care. IPHC
should use TCP_NODELTA, ECRTP should send absolute values, ROHC
should be adapted as discussed in [RFC4224]. An evaluation and
simulation of ECRTP and ROHC reordering is given in
[REORDER-EVAL].
Spencer (Probably a Nit): It wasn't obvious to me whether these
recommendations are sufficient to "implement and deploy
with care", or
whether additional precautions must be taken. Even putting these
recommendations in a numbered list immediately after
"deployed with care"
would be sufficient, if these recommendations are sufficient.
This goes back to a discussion with Allison Mankin RE CRTP issues
discussed icw RFC 4446. There was no further list of recommendations
out of that discussion, rather, the point is that in packet-lossy
environments, for example, CRTP may not work well and ECRTP
may perform
better. Some folks felt that CRTP should be excluded because of that
problem. There were, however, other concerns raised on
deploying ECRTP
(e.g., CRTP is already widely deployed, plus other reasons).
Spencer: would it be appropriate to say "implement and deploy
with care:",
and then put the recommendations in a numbered list? My
concern was pretty
basic - if I follow these recommendations, do I still have a
problem, or am OK?
There really isn't any list of 'recommendations' as to how to 'deploy
(CRTP) with care' in the case of lossy links and reordering issues, that
phrasing should be removed as misleading. Rather, we can further
explain the issue with CRTP, as described in [RFC3545]:
"5.4 Packet Reordering
...Although CRTP is
viewed as having risks for a number PW environments due to reordering
and loss, it is still the protocol of choice in many cases. CRTP was
designed for reliable point to point links with short delays. It
does
not perform well over links with high rate of packet loss, packet
reordering and long delays. In such cases ECRTP [RFC3545] may be
considered to increase robustness to both packet loss and misordering
between the compressor and the decompressor. This is achieved by
repeating updates and sending of absolute (uncompressed) values in
addition to delta values for selected context parameters. IPHC should
use ..."
Thanks again,
Regards,
Jerry
Again, thanks for a quick followup, while I can still
remember what I was
thinking when I wrote the review :-)
Spencer
_______________________________________________
Ietf mailing list
Ietf(_at_)ietf(_dot_)org
https://www1.ietf.org/mailman/listinfo/ietf