Re: [re-ECN] Two questions about CONEX

[Bob Briscoe <[e-mail]>]

Stewart,

At 15:46 10/05/2010, Stewart Bryant wrote:
>I would like to clarify what my concern is here.
>
>If the marking is in every packet then the forwarder needs to take 
>action on every packet.
>
>The thing that makes IP (and MPLS) cheap and fast is that we only do 
>the minimum amount of processing in forwarding and do as much as we 
>can by other means. In particular everything other than forwarding 
>happens at a much slower rate, is often asynchronous WRT forwarding, 
>and usually happens in a process that can schedule the processing at 
>some convenient time.

That was exactly the goal of re-ECN (which is a candidate protocol for ConEx).

The design of re-ECN is I think the same as the approach you loosely 
called 'OAM-based', then clarified below. Traffic control boxes take 
note of the per-packet re-ECN markings. But all other forwarding 
boxes ignore them (preferably marking ECN though, when congested).

The idea is to *remove* per-packet operations from forwarding boxes, 
shifting information from forwarding boxes to edge control boxes, and 
doing the traffic control once at the outermost edges of the internetwork.

One of our more ambitious goals was to be able to do all-optical 
packet forwarding at inter-domain borders (and within domains) with 
reduced photonic processing per packet. That requires pushing the 
incentives to control traffic back to the first ingress to the internetwork.

At the outer edge, the minimum per-packet policing function we have 
managed to achieve without weakening the incentives is a single token 
bucket per user - but unlike a traditional token bucket it regulates 
only the bit-rate of re-ECN marked packets, not all packets.


>I used the term OAM rather loosely earlier to indicate a packet that 
>traversed the path, fate sharing with the data, containing 
>information of note intended for the intermediate systems.  I make 
>no comment on how such packets might be carried or fate shared.
>
>My question here is that whilst I can see that we need to do better 
>than the 15mins that systems seem to do at the moment, do we need to 
>go all the way to making the reporting the congestion state on every packet.

This question might be born of complacency that it would be easy to 
stretch to a coarser timescale. Perhaps this stems from assuming 
congestion and utilisation are related. But the two are linked 
differently for different transport behaviours.

This is why LEDBAT-like transports can send a lot of data, helping to 
create high utilisation, but not be responsible for hardly any of the 
congestion. While other transports can contribute a lot to the 
congestion, without sending much data (e.g. unresponsive streaming).

That's why it's important that ECN markings (and drops) are 
per-packet. Also, by using in-band markings, you get a secure binding 
for free between the congestion information and the data. That buys a 
lot of simplicity.

I think worrying about over-precision is a red herring. For a traffic 
control box to add up the sizes of marked packets really isn't heavy 
processing.

If it's cheapest to get a precise result, why make the system 
complicated in order to make it imprecise?




Bob



>- Stewart
>
>
>
>On 10/05/2010 14:17, ken carlberg wrote:
>>just a couple of various thoughts....
>>
>>if there are alternatives to Conex, cool.  It would be nice to hear 
>>people volunteering for a design team and see what they come up 
>>with.  But i think that effort then probably needs to go to the 
>>IETF list to kick it off for wider distribution/advertisement, and 
>>then possibly have another list to go into the specifics and start 
>>separate BoF.
>>
>>I found that I agreed a fair amount what was previously sent to the 
>>list by Kevin Mason.  I'd add that if OAM was the direction to 
>>take, then isn't the direction of a solution now being pared down 
>>to the boundaries of a specific transit -- and really just the leaf 
>>transit?  My impression with Conex was that it was not constrained 
>>(even policy-wise) to a single transit but was independent to that 
>>because we are placing the info in the IP packet.  But please 
>>correct me if I'm missing something.
>>
>>-ken
>>
>>
>>
>>On May 10, 2010, at 8:23 AM, Woundy, Richard wrote:
>>
>>
>>>If folks in general (not just the IESG) think there are a lot of 
>>>alternatives to Conex, maybe the ADs should form a design team 
>>>that uses a real-time form of communication to talk this through? 
>>>I'm thinking any combination of f2f meeting, conference call, or 
>>>Webex to start...
>>>
>>>-- Rich
>>>
>>>________________________________
>>>
>>>From: Stewart Bryant [mailto:[e-mail]]
>>>Sent: Sat 5/8/2010 1:32 PM
>>>To: Woundy, Richard
>>>Cc: [e-mail]
>>>Subject: Re: [re-ECN] Two questions about CONEX
>>>
>>>
>>>
>>>On 08/05/2010 14:40, Woundy, Richard wrote:
>>>
>>>>So a couple of questions back to you.
>>>>
>>>>1. I gave the 'CMTS congestion management' as a representative 
>>>>example. But what if I need similar measurements/mechanisms at my 
>>>>backbone interconnects? (That's probably the second place I would 
>>>>need to worry about congestion.) Can these OAM messages make it there?
>>>>
>>>>
>>>That would depend on how they are carried. For example (and only for
>>>example) if they were embedded in the transport protocol itself they
>>>would go where ever the transport went.
>>>
>>>2. A typical subscriber host is behind one (or more) Ethernet/WiFi 
>>>home gateways, then behind a broadband modem that may or may not 
>>>use Ethernet as a layer two transport over the broadband access 
>>>network. So are we talking about a layer 2 or 3 OAM message? If 
>>>layer 3, how would a host behind a NAT know who to address it to? 
>>>(I don't personally like NAT66 but that may be the subscriber's 
>>>choice rather than mine.) If layer 2, how do I get the subscriber 
>>>to upgrade all of their home CPE? Plus how many layer 2 
>>>technologies need to define this OAM message? Or do we need to 
>>>define a new home network layer 2 (please no)?
>>>
>>>See above.
>>>
>>>>3. What if the OAM message needs to be originated by the 
>>>>application/content server side, rather than the subscriber side? 
>>>>How would that server know which CMTS / interconnect router to send it to?
>>>>
>>>>
>>>See above.
>>>
>>>>-- Rich
>>>>
>>>>P.S. The long time constant in the CMTS congestion management 
>>>>feedback loop is dependent on the state of the art today (DOCSIS 
>>>>specs require 15 minute IPDR polling cycle support for the CMTS). 
>>>>In some ways this is a bug, in some ways this is a feature. :)
>>>>
>>>>
>>>I agree, but I was not thinking of 15mins, on the other hand I was not
>>>thinking of packet rate which is where I believe current thinking is.
>>>
>>>- Stewart
>>>
>>>
>>>
>>>>________________________________
>>>>
>>>>From: [e-mail] on behalf of Stewart Bryant
>>>>Sent: Sat 5/8/2010 2:24 AM
>>>>To: [e-mail]
>>>>Subject: [re-ECN] Two questions about CONEX
>>>>
>>>>
>>>>
>>>>
>>>>As I was reading the recent email on CONEX I got the impression that a
>>>>lot of the interest in CONEX is to instrument the network so that
>>>>operators can identify the root causes of congestion and understand the
>>>>impact on the network in more detail.
>>>>
>>>>    I also get the impression that at least some of the operators are
>>>>looking for a relatively long time constant in the feedback loop.
>>>>
>>>>That causes me to wonder where CONEX fits in relative to IPFIX which is
>>>>a mechanism that is designed to monitor the flows in a network and
>>>>report this information to the network operator.
>>>>
>>>>As I noted in a earlier thread, I am also interested in understanding
>>>>why the host needs to use the data packets themselves to indicate the
>>>>expected congestion state to the network rather than using a fate
>>>>sharing OAM mechanism?
>>>>
>>>>- Stewart
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>_______________________________________________
>>>>re-ECN mailing list
>>>>[e-mail]
>>>>https://www.ietf.org/mailman/listinfo/re-ecn
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>
>>>--
>>>For corporate legal information go to:
>>>
>>>http://www.cisco.com/web/about/doing_business/legal/cri/index.html
>>>
>>>
>>>
>>>
>>>_______________________________________________
>>>re-ECN mailing list
>>>[e-mail]
>>>https://www.ietf.org/mailman/listinfo/re-ecn
>>>
>>
>>
>
>
>--
>For corporate legal information go to:
>
>http://www.cisco.com/web/about/doing_business/legal/cri/index.html
>
>
>_______________________________________________
>re-ECN mailing list
>[e-mail]
>https://www.ietf.org/mailman/listinfo/re-ecn

________________________________________________________________
Bob Briscoe,                                BT Innovate & Design