Re: [Int-area] draft-bonica-intarea-gre-mtu and ECN

[Bob Briscoe <[e-mail]>]

Lars, Ron,

At 15:20 07/08/2013, Eggert, Lars wrote:
>Hi,
>
>On Aug 6, 2013, at 19:34, Ronald Bonica <[e-mail]> wrote:
> > Section 5.3 of RFC 3168 specifies procedures for handling the ECN 
> bit when reassembling fragmented packets. These rules must be 
> observed by any device that reassembles fragmented packets, 
> including tunnel egress routers. It would be reasonable to make 
> note of this in draft-bonica-intarea-gre-mtu.
>
>agreed.

+1

>You should also talk about what happens with the DSCP.

It is possible that different fragments of a packet will get 
different DSCPs (e.g. if some packets are re-marked due to being out 
of contract in a DS policer). However, RFC2474, RFC2983 (Diffserv & 
Tunnels) and RFC2597 (Diffserv AF) all fail to mention anything about 
what to do on re-assembly.

So I wouldn't open this can of worms, given no-one else has. It 
probably doesn't matter greatly which of the DSCPs on the fragments 
ends up being used for the reassembled packet, but I guess most 
implementations will use the DSCP on the first fragment, even if they 
haven't consciously thought about it.

It wouldn't take much to update 2474 to just say that - if someone 
has some spare clock cycles to waste.



> > Section 9.1 of RFC 3168 as well as RFC 6040, specify procedures 
> for propagating ECN bits between the tunnel payload and delivery 
> header. These rules apply to all tunneled packets, regardless of 
> whether they are fragmented. Because the scope of 
> draft-bonica-intarea-gre-mtu is limited to MTU and fragmentation 
> issues, a discussion of these rules seems to be beyond the scope of 
> draft-bonica-intarea-gre-mtu.
>
>Agreed.

+1

>However, given that the GRE spec predates the ECN spec, what are GRE 
>implementations actually doing with the ECN/DSCP fields? 
>Unfortunately, 3168 did not update 2784, although GRE can also 
>provide an IP-in-IP tunnel type.

Yes, I would be interested in an answer to that. Particularly given 
recent ideas on using GRE tunnels in data centres to turn on ECN 
edge-to-edge. See draft-briscoe-conex-data-centre-01 or 
<http://www.ietf.org/proceedings/87/slides/slides-87-conex-4.pdf>

PS. We also have an individual draft that clarifies that the RFC6040 
propagation of ECN should be used by all the different tunnelling 
protocols, including GRE: 
<http://tools.ietf.org/html/draft-briscoe-tsvwg-ecn-encap-guidelines-02>


Bob


>Lars
>
>PS: CC'ing Bob, who also commented.
>

________________________________________________________________
Bob Briscoe,                                                  BT