Stewart I proposed precisely this approach at IETF-80 (unfortunately, since becoming AD you haven't been able to attend TICTOC meetings :( )
The WG decided to proceed with the simpler approach first, and to consider pursuing the more generic approach after the first one progressed. I call the present draft "simpler" because it reuses existing encapsulations and 1588 procedures. It just requires control protocol work and having the LSR recognize a specific label. The idea is that the specific label triggers some specific processing, just like specific MAC addresses or IP addresses trigger specific functionality in switches or routers respectively. (I agree that it is not just a matter of sending the packet up to a CPU for special processing, since it requires a HW timestamp, but there are several approaches to solving this problem.) We originally spoke about using a reserved label, but feedback from the MPLS WG was that we had little chance of getting such a limited resource. So the obvious choice was telling the LSR about specific labels. Since Beijing we have had regular participation of people from the MPLS WG (including George) and some of the TICTOC regulars, including the draft authors and yours truly, participate in MPLS too. Y(J)S -----Original Message----- From: Stewart Bryant [mailto:[email protected]] Sent: Sunday, November 27, 2011 02:13 To: Yaakov Stein Cc: Shahram Davari; '[email protected]' Subject: Re: [TICTOC] FW: I-D Action: draft-ietf-tictoc-1588overmpls-02.txt Yaakov I was not thinking of a particularly radical packet type. A label stack, a 64 bit delay field, and then any timing payload you like (specific details out of scope for the draft). The P routers then just add and subtract the current time from the delay field to update it with the dwell time. The Edge will know what the timing payload type is and can make the TC correction based on the delay field. The important point is that this works for any packet type that needs to know delay, be that a timing payload such as 1588v2, 1588v3, NTPv4, NTPv5 etc etc. It also as Yaakov notes works for any other payload type such as a payload that needs to perform synchronous delivery at multiple disjoint endpoints, or a 1+1 protection system that is required to provide synchronied streams etc etc. - Stewart On 26/11/2011 20:24, Yaakov Stein wrote: > Shahram and Stewart > > If we need intermediate MPLS nodes to perform special processing on 1588oMPLS > packets > there are several methods to lower the processing requirements. > Of course, DPI could be performed to go below the MPLS and IP headers as > Shahram said, > but as Stewart pointed out this would be prohibitively expensive. > > Two methods have been proposed. > The method of the present draft is to use the standard encapsulations > (after ensuring that 1588 is supported) > and to inform the intermediate nodes that the particular label value being > used is special. > For this special label value the node has been informed of what to do, > e.g., has the offset of a TC. > Any use of TC is necessarily a layer violation > (after all, the timestamp is a layer-0 entity and we are placing it in a > layer 2 or higher field), > but correcting a field inside 1588 in UDP in IP in MPLS > is not really that much worse than correcting on in 1588 in UDP in IP in > Ethernet. > > The alternative method that I proposed is to invent a completely new > timestamping mechanism. > This has the advantage of being applicable to all MPLS packets > (and thus can solve other problems), > but requires inventing yet another timing distribution protocol. > I know that Stewart succeeded in inventing a new packet loss and delay > measurement protocol for TP, > but I didn't gauge support in TICTOC for something new here. > > Y(J)S > > -----Original Message----- > From: Stewart Bryant [mailto:[email protected]] > Sent: Thursday, November 24, 2011 19:30 > To: Shahram Davari > Cc: '[email protected]'; '[email protected]' > Subject: Re: [TICTOC] FW: I-D Action: draft-ietf-tictoc-1588overmpls-02.txt > > Shahram > > I will ponder the answer to this question, but will note that > you have not addressed my second question which relates > to whether there is MPLS WG buy-in for this proposal. > > - Stewart > > > > On 24/11/2011 16:34, Shahram Davari wrote: >> Hi Stewart, >> >> The parsing required by the draft is not complex and almost all MPLS routers >> have support it already. The idea was to reuse existing data plane >> mechanisms and not invent a new one. This I believe is in the spirit of IETF >> to reuse existing mechanisms. >> >> I don't believe adding a shim makes the design simpler. You still need to >> detect that such shim exists and for that you need parsing that doesn't even >> exist today. >> >> >> This draft has been implemented by vendors, so we have a working code and I >> believe we also have rough consensus. >> >> Thanks >> Shahram >> >> >> >> ----- Original Message ----- >> From: Stewart Bryant [mailto:[email protected]] >> Sent: Thursday, November 24, 2011 07:58 AM >> To: [email protected]<[email protected]> >> Cc: [email protected]<[email protected]>; >> [email protected]<[email protected]> >> Subject: Re: [TICTOC] FW: I-D Action: draft-ietf-tictoc-1588overmpls-02.txt >> >> Can we wind back to my original points here >> which have not addressed. >> >> Why are is the WG proposing a design that >> needs such complex parsing, against the >> ethos of MPLS, when a simpler design >> would be more universally applicable? >> >> Does the WG have any input to suggest that >> the design will survive a review by MPLS/PWE3 >> WG and then by IESG? >> >> - Stewart >> >> >> On 22/11/2011 09:12, Stewart Bryant wrote: >>> Speaking as an individual here, I really have a hard time >>> understanding why it is necessary to have quite the >>> egregious layer violation that this draft uses. >>> >>> The idea of having an LSP type that is dedicated to >>> tracking the time of passage through the network >>> is a good idea. However MPLS is completely geared >>> to the concept that only the LSP endpoints know >>> how to resolve the payload type. >>> >>> The function that you require could be achieved >>> by including a shim that contains the >>> time compensation information and adjust the >>> payload on egress from the LSP. That would be >>> rather more consistent with the MPLS architecture. >>> >>> I have not seen a request for review by the MPLS >>> or PWE3 WGs and I would suggest that you request >>> that sooner rather than later since it is inevitable >>> that the draft will be sent there later in it's life, and >>> if they do not subscribe to your mode of operation >>> the draft is unlikely to progress. >>> >>> I would also suggest that you discuss the extent >>> of layer violation with your AD to make sure he is >>> confident that this draft will pass IESG review. >>> >>> - Stewart >>> >>> >>> _______________________________________________ >>> TICTOC mailing list >>> [email protected] >>> https://www.ietf.org/mailman/listinfo/tictoc >>> > -- For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/index.html _______________________________________________ TICTOC mailing list [email protected] https://www.ietf.org/mailman/listinfo/tictoc
