Tal,
I think that we are in full agreement regarding a single-label GAL-only stack:
It allows the LSRs to modify PTP packets (as TC must do), but it does not
support forwarding (unicast and multicast). Hence it is useless.
IMO GAL/G-ACH are a clear dead end in this discussion.
If you have been following this list, I've suggested allocating a new reserved
label which, at the top of the stack, would indicate PTP packets.
IMHO this is the only method that would somehow accommodate TC (which is a
layer violation, of course) without completely breaking MPLS data plane.
However, this approach did not gain support on the list.
"PTP FEC" IMO is also a dead end. E.g., you would not be able such popular
recovery techniques as Facility FRR for this FEC, etc.
At the same time, if TC support is not required, there is no need to invent any
special encapsulation of PTP over MPLS. All the rest of the use cases can be
easily accommodated with PTP/UDP/IP/MPLS without inventing new wheels.
My 2c,
Sasha
From: Tal Mizrahi [mailto:[email protected]]
Sent: Monday, July 19, 2010 5:55 PM
To: Alexander Vainshtein
Cc: [email protected]
Subject: Re: [TICTOC] FW: 1588 over MPLS draft
Hi Sasha, all,
A couple of comments:
1. Generally, regarding using GAL+ACH: according to RFC 5586: "LSRs MUST NOT
modify the G-ACh message, the ACH or the GAL towards the targeted destination".
However, a PTP capable LSR that functions as a TC must modify the packet
(including the correctionField and the UDP checksum). This means that if an LSR
functions as a TC, either (a) the functionality in RFC 5586 must be enhanced to
allow modification, or (b) the LSR terminates all incoming PTP messages, and
then re-generates them, which may burden the control plane.
2. Sasha, regarding your option 1 below: if each packet has a label stack with
1 label (GAL), it raises the question how to distinguish between single-target
and multi-target PTP packets. PTP calls for both multicast frames (e.g.), and
unicast (e.g. Delay_Req).
That's not to say I am against using the GAL, but it needs some further
refinement.
Tal.
-----Original Message-----
Yaakov and all,
Please note that GAL and, by implication, the ACH header are only looked at by
an LSTR if GAL happens to be the top label in the stack (either because it has
been the top label in originally received packet, or because all the labels
above it have been popped by this LSR).
This leaves two options IMO:
1. You carry PTP directly across physical links using labeled packets that
have a label stack of depth 1 containing GAL.
In this case you can probably do want you want with the PTP packets'
payload, (e.g., support TC), but you need some new mechanism for forwarding
PTP packets along a multi-hop path.
2. You can carry PTP across any LSP using GAL at the bottom of the label stack.
In this case only the tail end of the LSP will be PTP-aware, i.e., TC will
not be supported with this option.
I suggest taking a look at
http://datatracker.ietf.org/doc/draft-ietf-mpls-tp-data-plane for details
regarding MPLS (and MPLS-TP) data plane.
Regards,
Sasha
-----Original Message-----
From: tictoc-bounces at ietf.org [mailto:tictoc-bounces at ietf.org] On Behalf
Of Yaakov Stein
Sent: Monday, July 19, 2010 12:06 PM
To: Mikael Abrahamsson
Cc: tictoc at ietf.org
Subject: Re: [TICTOC] FW: 1588 over MPLS draft
No. Not a new Ethertype - we are talking about MPLS NOT Ethernet.
There is a protocol type (it's actually called a "channel type")
in the Ach control word. See RFC 4385.
Right now only a few are defined (raw BFD, IPv4, IPv6).
Y(J)S
-----Original Message-----
From: Mikael Abrahamsson [mailto:swmike at swm.pp.se]
Sent: Monday, July 19, 2010 09:48
To: Yaakov Stein
Cc: stbryant at cisco.com; tictoc at ietf.org
Subject: Re: [TICTOC] FW: 1588 over MPLS draft
On Sun, 18 Jul 2010, Yaakov Stein wrote:
> 1) define a new protocol type (plenty of openings in THAT registry!)
New protocol type on what level? Ethernet, so this would involve a new
ethertype?
If routers generally can look that far into the packet on the correct
forwarding level (I doubt it though) then that would be the least
intrusive, but having LSRs look for ethertype within MPLS labeled packets
sounds kind of advanced to do that early in the receive path?
Why not do it more like an MPLS L3 VPN terminated/routed by all
involved&&aware routers, then it would signal special labels to its
neighbours that would be local significance only? But now we're talking
handling it like a tree and that would involve routing protocols as
well... Basically this would be like multicast IP and could leverage all
the multicast MPLS standards out there.
--
Mikael Abrahamsson email: swmike at swm.pp.se
_______________________________________________
TICTOC mailing list
TICTOC at ietf.org
https://www.ietf.org/mailman/listinfo/tictoc
_______________________________________________
TICTOC mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/tictoc
