Re: [PCN] Encoding Option: '01' & '10' Encoding

[<[e-mail]>]

Kwok

Just wanted to clarify what your proposal is. I think it's:

PCN Meaning		ECN 	 DSCP

Admission Mark	00	 PCN 1
PCN-capable   	01	 PCN 1
PCN-capable   	10	 PCN 1
Termination Mark	11	 PCN 1

I think something like this is an option worth considering. Are these
the pros & cons?


Pros:
o  It only needs one DSCP, per PCN behaviour aggregate. In normal
circumstances where the PCN-domain is configured correctly, then
PCN-traffic is distinguished from non-PCN traffic by use of this DSCP.

o  For a router to classify a packet as PCN, it simply reads its DSCP.
This is probably easier than checking both its DSCP & ECN fields. 

Unclear:
o  If the PCN-boundary-nodes are misconfigured so that PCN packets can
sneak out of the PCN-domain, or non PCN-packets can sneak in, then it
needs more analysis.
(Kwok, your scheme is quite good from this point of view. '00' sneaking
out?)

Cons
 o  Termination marking and admission marking are not independent -
before a router writes TM is must check that the packet isn't already
admission marked. (Note: Over-writing a TM with AM is not allowed for
3SM & CL, see draft-charny-pcn-comparison.)

o  It needs an extra DSCP, per PCN behaviour aggregate. 

o  If a packet (in a PCN flow) arrives at the egress with its ECN = 01,
10 or 11 and it really is required to preserve this info across the
PCN-domain, then some complexity is needed at the PCN-boundary-nodes
(tunnelling)


Hope you're de-fevered!
phil

> -----Original Message-----
> From: [e-mail] [mailto:[e-mail]] On Behalf Of
> Kwok-Ho Chan
> Sent: 22 February 2008 19:08
> To: [e-mail]
> Subject: [PCN] Encoding Option: '01' & '10' Encoding
> 
> Hi all:
> Following up on Toby's E-Mail, I would like to bring out the possible
> choices for
> usage of '01' and '10' bit patterns (of the 2 bit ECN field in the IP
> Header) for PCN.
> 
> For ECN:  '01' = ECT(1), '10' = ECT(0).  Meaning Not Congested, and
ECN
> Capable
> Transport, and ECN nonce usage.
> 
> Toby's encoding suggest the NOT using of '01' and '10'.
> It is a good suggestion if we have more bits to use.  But we don't,
and
> the use
> of additional DSCPs have other affects and trade-offs.
> 
> Alternative is to use both '01' and '10' to indicate the packet is PCN
> capable
> (in another words belong to a PCN capable flow).
> 
> The benefit of using '01' and '10' is allowing it to be used for PCN
> purpose.
> 
> We have agreed to use a DSCP to indicate PCN Capable packets (flows),
> hence the PCN Capable notion (encoding state) is taken care of.
> 
> We still need to represent Not Pre-Congested notion (encoding state).
> IMHO, the '01' and '10' bit patterns are good choices for encoding the
> Not Pre-Congested encoding state.
> 
> Lets visit the benefits of such use:
> 1. Allow the representation of a PCN encoding state using ECN bits.
>      Minimizing the use of additional DSCPs.
> 
> 2. If a PCN '01' or '10' encoded packet receives ECN router treatment:
>      The ECN router will not change the encoding if it does not
experience
>      congestion.
>      The ECN router will change the encoding to '11' (CE) if it
> experienced
>      congestion, trusting the offered load to the congestion point
> will be reduced.
>      If a PCN Edge Node receives the '11' encoding and if we choose it
to
> mean
>      Termination, then the offered load to the congestion point will
> be reduced.
>      PCN terminates the flow, hence this should be ECN (at least ECN
when
>      used for TCP) friendly.  Detailed discussions of how friendly
should
> be
>      an algorithm matter and will need to involve algorithm
> comparisons with ECN.
>      If an ECN End Node receives the '11' encoding, it will mean
interpret
> it
>      as CE and will reduce the offered load as describe by RFC 3168.
>      How friendly is this ECN flow to the PCN domain is a discussion
> of algorithms
>      and will have to make assumptions of using ECN for TCP and which
> PCN algorithm.
>      The how friendly discussions should be part of the algorithm
> discussions.
> 
> 3. When ECN packets pass through PCN Interior Nodes with no Pre-
> Congestions,
>      no affects to ECN and PCN.  This includes ECN packets using ECN
> nonce.
> 
> 4. When PCN packets pass through ECN routers with no Congestions,
>      no affects to ECN and PCN.
> 
> 5. If we don't have to worry about ECN nonce passing through PCN
nodes,
>      we can use one of the two bit patterns for another PCN encoding
> state.
> 
> Point # 2 above discusses:
> 1. ECN packets entering PCN domain and then leaving the PCN domain
>      back into ECN routers and endpoints.
> 2. PCN packets leaving the PCN domain and encounters ECN routers and
>      PCN endpoints.  Does it make sense to talk about PCN packets
> encountering
>      ECN (TCP) endpoints?  If the ECN endpoint handles this PCN
packet,
> then
>      it should reduce the flows offered load, the functionality PCN
> Interior Nodes expects.
> 
> Based on the assumption of using PCN DSCP to separate PCN traffic and
> non PCN traffic:
> - When we talk about ECN packet entering PCN domain, we mean somehow
>    the ECN packet uses the PCN DSCP.  There are DiffServ methods to
> prevent this.
> - When we talk about PCN packet leaving PCN domain, we mean somehow
>    the PCN packet encounters routers that does not understand the PCN
DSCP
>    and interpret the PCN encoding differently.
> 
> The bad part of such use:
> 1. Making the PCN algorithms to handle how friendly they are with ECN
> algorithm.
> 
> Please notice the information provided above are in:
>
http://www.ietf.org/internet-drafts/draft-chan-pcn-encoding-comparison-
> 02.txt
> 
> I am adding the details of Toby's suggestions to the -03 of the draft.
> 
> Please let us know if there are points or concerns not addressed.
> Thanks!
> 
> Sorry that I have been sick with the flu (fever, aches, cough, etc)
> the past few days.
> Just got below 100 degree temperature with my fever this morning.
> 
> -- Kwok --
> 
> _______________________________________________
> PCN mailing list
> [e-mail]
> http://www.ietf.org/mailman/listinfo/pcn
_______________________________________________
PCN mailing list
[e-mail]
http://www.ietf.org/mailman/listinfo/pcn