Joe,
Following up my email to you in May quoted further down, you made me
realize that RFC6040 did not address what to do with ECN during
fragmentation and reassembly. So I've added the following to my local
copy of draft-ietf-tsvwg-rfc6040-update-shim (to be posted later today),
which recently went through TSVWG last call, and will imminently be last
called on various int-area lists, I believe.
These are quite significant updates to outer fragment processing at the
tunnel egress. But, given something has to be said, I can't think of a
better way (see the original quoted email about why the logical OR of
the ECN codepoints as defined in RFC3168 is no longer sufficient - and
it's no simpler anyway).
5. ECN Propagation and Fragmentation/Reassembly
The following requirements update RFC6040, which omitted handling of
the ECN field during fragmentation or reassembly. These changes
might alter how many ECN-marked packets are propagated by a tunnel
that fragments packets, but this would not raise any backward
compatibility issues:
If a tunnel ingress fragments a packet, it MUST set the outer ECN
field of all the fragments to the same value as it would have set if
it had not fragmented the packet.
As a tunnel egress reassembles sets of outer fragments
[I-D.ietf-intarea-tunnels] into packets, it SHOULD propagate CE
markings on the basis that a congestion indication on a packet
applies to all the octets in the packet. On average, a tunnel egress
SHOULD approximately preserve the number of CE-marked and ECT(1)-
marked octets arriving and leaving (counting the size of inner
headers, but not encapsulating headers that are being stripped).
This process proceeds irrespective of the addresses on the inner
headers.
Even if only enough incoming CE-marked octets have arrived for part
of the departing packet, the next departing packet SHOULD be
immediately CE-marked. This ensures that CE-markings are propagated
immediately, rather than held back waiting for more incoming CE-
marked octets. Once there are no outstanding CE-marked octets, if
only enough incoming ECT(1)-marked octets have arrived for part of
the departing packet, the next departing packet SHOULD be immediately
marked ECT(1).
For instance, an algorithm for marking departing packets could
maintain a pair of counters, the first representing the balance of
arriving CE-marked octets minus departing CE-marked octets and the
second representing a similar balance of ECT(1)-marked octets. The
algorithm:
o adds the size of every CE-marked or ECT(1)-marked packet that
arrives to the appropriate counter;
o if the CE counter is positive, it CE-marks the next packet to
depart and subtracts its size from the CE counter;
o if the CE counter is negative but the ECT(1) counter is positive,
it marks the next packet to depart as ECT(1) and subtracts its
size from the ECT((1) counter;
o (the previous two steps will often leave a negative remainder in
the counters, which is deliberate);
o if neither counter is positive, it marks the next packet to depart
as ECT(0);
o until all the fragments of a packet have arrived, it does not
commit any updates to the counters so that, if reassembly fails
and the partly reassembled packet has to be discarded, none of the
discarded fragments will have updated any of the counters.
During reassembly of outer fragments [I-D.ietf-intarea-tunnels], if
the ECN fields of the outer headers being reassembled into a single
packet consist of a mixture of Not-ECT and other ECN codepoints, the
packet MUST be discarded.
A tunnel end-point that claims to support the present specification
MUST NOT use an approach that results in a significantly different
ECN-marking outcome to that defined by the "SHOULD" statements
throughout this section. "SHOULD" is only used to allow similar
perhaps more efficient approaches that result in approximately the
same outcome.
Bob
On 16/05/2019 22:14, Bob Briscoe wrote:
Joe,
Sorry I missed this posting at the time (my mail filters moved both
cross-postings into my int-area box which I check only rarely).
On 27/04/2019 18:13, Joe Touch wrote:
Cross-posting to let both communities know:
- it would be useful for these documents to address how fragmentation
and reassembly affects these signals
(esp. if reassembling fragments with different ECN values)
[BB] This is addressed by the re-framing section
<https://tools.ietf.org/html/draft-ietf-tsvwg-ecn-encap-guidelines-12#section-4.6>
in ecn-encap-guidelines, altho it doesn't give examples of what might
have caused frame boundary misalignment, so fragmentation is not
specifically mentioned. I think I will add an explicit mention of
fragmentation (if only so a search finds that section).
Actually I've realized that this highlights an inconsistency between
the advice on ECN and fragment reassembly in RFC3168 and in
ecn-encap-guidelines.:
* RFC3168 requires that the ECN marking of a reassembled packet is
the logical OR of the ECN marks on the fragments,
* whereas ecn-encap-guidelines recommends marking the same number of
outgoing as incoming octets when reassembling L2 frames or
tunnelled packets with different boundaries - using a simple
counter to track the balance.
In fact, it was the review of RFC3168 by me and Jon Crowcroft back in
2001 that originally raised the question of how to handle reassembly
of ECN-marked fragments
<https://tools.ietf.org/html/draft-ietf-tsvwg-ecn-ip-00#section-11>.
I'll quote a passage from the review, which I think justifies the
recommendation in ecn-encap-guidelines to count marked bytes, rather
than use the logical OR of RFC3168:
To use the logical OR of the marking of all fragments might be a pragmatic
solution, particularly for congestion control protocols like TCP where one
loss per round trip is treated identically to many. However, it is becoming
more common to see large numbers of packets per round trip time as data
rates increase while packet sizes and the speed of light haven't increased
for many years. Therefore it is to be expected that newer congestion
control protocols might take more accurate account of the number of packets
marked in a round trip. Hence, the inaccuracy of a logical OR during
re-assembly at the IP layer is best avoided.
I'm not too worried about the inaccuracy of using a logical OR, but I
think it best to recommend more accurate and no less costly counting.
The only justification for the logical OR was that TCP only reacted to
one ECN mark per RTT. But that is changing now, and the behaviour of
one transport should not be embedded in lower layers anyway.
- it would be useful for these documents to consider
draft-ietf-intarea-tunnels (which relates to the above) and its
discussion on many of the protocols cited
I can't find anything in draft-ietf-intarea-tunnels that ought to be
cited from ecn-encap-guidelines or rfc6040-update-shim. Did you have
something specific in mind?
I do want to raise a question about the following sentence, which
precedes the mention of ECN:
There are exceptions to this rule that are explicitly intended to
relay signals from inside the tunnel to the network outside the
tunnel, typically relevant only when the tunnel network N and the
outer network M use the same network.
Was that last word meant to say "network protocol"?
Then, if that is what you meant, I would disagree. Many different
network protocols include concepts similar to Diffserv and/or ECN
(e.g. IEEE802.1p, MPLS and TRILL support both, etc), and there's
rarely a reason /not/ to propagate the concept between different
network protocols when they encapsulate each other, even tho it's not
always straightforward to do so.
Bob
Bob
Joe
On Apr 26, 2019, at 1:50 PM, Black, David <[email protected]
<mailto:[email protected]>> wrote:
This may be of interest to INT folks who are interested in tunnels
and encapsulations.
Comments by the WGLC deadline are encouraged, but comments after the
deadline are ok, as they’d have to be dealt with anyway at IETF Last
Call.
Thanks, --David
*From:*tsvwg <[email protected]
<mailto:[email protected]>>*On Behalf Of*Black, David
*Sent:*Wednesday, April 17, 2019 2:51 PM
*To:*[email protected] <mailto:[email protected]>
*Subject:*[tsvwg] 2nd WGLC on ecn-encap-guidelines and
rfc6040-update-shim drafts, closes 6 May 2019
[EXTERNAL EMAIL]
This email announces a 2nd TSVWG Working Group Last Call (WGLC) on
two drafts:
[1] Guidelines for Adding Congestion Notification to Protocols that
Encapsulate IP
draft-ietf-tsvwg-ecn-encap-guidelines-12
https://datatracker.ietf.org/doc/draft-ietf-tsvwg-ecn-encap-guidelines/
This draft is intended to become a Best Current Practice RFC
[2] Propagating Explicit Congestion Notification Across IP Tunnel
Headers
Separated by a Shim
draft-ietf-tsvwg-rfc6040update-shim-08
https://datatracker.ietf.org/doc/draft-ietf-tsvwg-rfc6040update-shim/
This draft is intended to become a Proposed Standard RFC.
This WGLC will run through the end of the day on Monday, May 6, 2019.
Comments should be sent to [email protected]
<mailto:[email protected]>list, although purely
editorial comments may be sent directly to the author. Please cc: the
WG chairs [email protected] <mailto:[email protected]> if
you would like the chairs to
track such editorial comments as part of the WGLC process.
No IPR disclosures have been submitted directly on either draft
Thanks,
David, Gorry and Wes
(TSVWG Co-Chairs)
_______________________________________________
Int-area mailing list
[email protected] <mailto:[email protected]>
https://www.ietf.org/mailman/listinfo/int-area
--
________________________________________________________________
Bob Briscoehttp://bobbriscoe.net/
--
________________________________________________________________
Bob Briscoe http://bobbriscoe.net/
_______________________________________________
Int-area mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/int-area
