Re: [mpls] [mpls-tp] [PWE3] 1588 over MPLS draft

[Ben Niven-Jenkins <[e-mail]>]

Sharam,

I am not sure there is a requirement for all devices to be upgraded. My thinking is as follows:

We have 3 types of 1588 MPLS devices:
1) Those that do not support 1588
2) Those that can terminate 1588 but do not support boundary clock
3) Those that can terminate 1588 and do support boundary clock.

For the purpose of this discussion (2) & (3) can probably be lumped together.

The requirement (to avoid more complex hardware processing) is that 1588 packets arrive at device types (2) & (3) with a label out of a "1588 label range". A label from the 1588 label range has no special meaning with respect to forwarding behaviour it simply carries an extra semantic that indicates that the packet is a 1588 packet, such a semantic is only required if the LSP is terminating on device types (2) & (3) or transiting device type (3).

Therefore the 1588 LSP does not need to use a label from a 1588 specific label range at every hop in the path, it only needs such a label as the incoming label to device types (2) & (3).

If labels are downstream assigned then the downstream device provides the label mapping for the incoming label for itself and so device types (2) & (3) can assign labels out of their 1588 label range, there is no requirement for the upstream device to realise such a label has an additional semantic and therefore there is no need to upgrade devices except those directly participating in the 1588 session (either as terminating devices or as boundary devices).

If labels are upstream assigned then the upstream device will need to assign a label from the 1588 label range and so the upstream device, even if it is of type (1), needs to realise it has to assign a label with 1588 semantics. The upstream (as well as downstream) device will require upgrading although context labels may provide some level of support already (I would have to think more about the applicability of context labels to this problem space).

The key thing to remember is that the semantic that the LSP is a 1588 LSP needs to be carried End to End when signalling the LSP but does not need to exist within the forwarding machinery for each device in the path unless that device is participating in the 1588 session in some way (including upstream allocation of 1588 labels).

Ben



On 15 Jul 2010, at 21:37, Shahram Davari wrote:

> Hi Joel,
> 
> OK, do you have a better proposal?
> 
> Thx
> Shahram
> 
> -----Original Message-----
> From: Joel M. Halpern [mailto:[e-mail]] 
> Sent: Thursday, July 15, 2010 1:32 PM
> To: Shahram Davari
> Cc: [e-mail]; [e-mail]; [e-mail]; [e-mail]
> Subject: Re: [mpls] [PWE3] 1588 over MPLS draft
> 
> I am sorry, but requiring every LSR in the infrastructure to get a 
> software upgrade in order to allow for the presence of 1588-supporting 
> devices is a MAJOR issue.  It is NOT backwards compatible.  Operators 
> (wether service provider or enterprise IT) do NOT like being told they 
> ahve to upgrade the software on all devices in their network in order to 
> add a capability that those devices are not engaging in.
> 
> While one can argue about the "ease" of the specific software change, 
> there is from an operational perspective no such thing as an easy 
> upgrade of any device, and there is definitely no such thing as an easy 
> upgrade of all the devices in the network.
> 
> Yours,
> Joel
> 
> Shahram Davari wrote:
>> Hi Joel,
>> 
>> There are 2 aspects to this puzzle:
>> 
>> 1) Hardware support for 1588 in a device. 
>> 2) Software support for 1588 in the form of allocating labels to PTP LSPs
>> 
>> A router that wants to perform 1588 time stamping must have both (1) and (2) functionality. While a router in 
>> the path of a PTP LSP that does not support (1) should support (2) and 
> advertise labels from some arbitrary
>> range for PTP LSP.
>> 
>> It is quite easy to upgrade a router software so that when it receives RSVP-TE label request to assign label 
>> from an arbitrary local range.
>> 
>> So I don't see any issue here.
>> 
>> Thanks,
>> Shahram
>> 
>> 
>> 
>> -----Original Message-----
>> From: Joel M. Halpern [mailto:[e-mail]] 
>> Sent: Thursday, July 15, 2010 12:06 PM
>> To: Shahram Davari
>> Cc: Mach Chen; S. Davari; Jia HE; [e-mail]; [e-mail]; [e-mail]; [e-mail]
>> Subject: Re: [mpls] [PWE3] 1588 over MPLS draft
>> 
>> Architecturally, as I understand the draft, the downstream neighbor of a 
>> 1588 supporting device (on the PTP carrying LSP) may not itself be a 
>> 1588 supporting device.  Is that correct?  If that is not correct, if 
>> instead the structure is that PTP supporting LSPs are established 
>> between adjacent enhanced devices, then I can see how some sort of a 
>> label range might make sense.
>> 
>> If however, the downstream node you refer to may be an existing LSR, 
>> then I do not see how you can ask an existing LSR to allocate a label 
>> range for a function it does not understand.
>> 
>> Yours,
>> Joel
>> 
>> Shahram Davari wrote:
>>> Hi Joel,
>>> 
>>> I agree with your logic, but the idea is a bit different:
>>> 
>>> All we need is a label range, it does not need to be globally unique in the network and it does
>>> not even need to be the same on the Tx and Rx direction of the same 
>> link. Although we need time
>>> stamping both on Rx and on Tx, but on Rx a node is in control of its 
>> own label and on Tx the
>>> downstream node should advertise some label range. What we don't want 
>> is that the downstream
>>> node to advertise random labels for each LSP carrying PTP.
>>> I will update the draft to mention that the label range does not require to be global but it can
>>> be per-platform or even per-interface.
>>> Thanks,
>>> Shahram
>>> 
>>> -----Original Message-----
>>> From: Joel M. Halpern [mailto:[e-mail]] 
>>> Sent: Wednesday, July 14, 2010 8:21 PM
>>> To: Mach Chen
>>> Cc: Shahram Davari; S. Davari; Jia HE; [e-mail]; [e-mail]; [e-mail]; [e-mail]
>>> Subject: Re: [mpls] [PWE3] 1588 over MPLS draft
>>> 
>>> Remember that if what is required is that the message arriving at the 
>>> 1588 supporting device have a certain label range, that is a purely 
>>> local matter.  As long as the singaling carries the indication that the 
>>> LSP is to be dedicated to 1588 traffic (a reasonable extension to the 
>>> signaling), the downstream switch can pick what label to assign to that LSP.
>>> 
>>> If the goal is to have the outgoing label be from a specific range, that 
>>> is essentially impossible in the proposed architecture.  The proposed 
>>> architecture is one in which 1588 switches are peers with existing LSRs 
>>> using MPLS.  As such, in order to have a label range for outgoing 
>>> labels, the existing MPLS LSR would somehow have to know about this 
>>> reserved label range, not use those labels for other purposes, and 
>>> assign those labels to 1588 LSPs.  All of which is 
>>> non-backwards-compatible.  If the 1588 LSPs were actually pseudowires, 
>>> tunneled over LSPs, then any number of things are possible.
>>> 
>>> Yours,
>>> Joel M. Halpern
>>> 
>>> Mach Chen wrote:
>>>> Hi Shahram,
>>>> 
>>>> I agree that the label range idea is an efficient way that could reduce 
>>>> the storage requirement of PHY chips.
>>>> 
>>>> It's easy to require one or limited nodes to reserve a label range, but 
>>>> IMHO, it very difficut to require all nodes of a large network to do 
>>>> this and even worse when some labels are already used by other LSPs, 
>>>> unless there are some mechanisms to negotiate/advertise(e.g., flooding 
>>>> the label range by IGP within the network) the proper label range hence 
>>>> to aviod label collision.
>>>> 
>>>> Best regards,
>>>> Mach
>>>> 
>>>> --------------------------------------------------
>>>> From: "Shahram Davari" <[e-mail]>
>>>> Sent: Thursday, July 15, 2010 1:58 AM
>>>> To: "Mach Chen" <[e-mail]>; "S. Davari" <[e-mail]>; 
>>>> "Jia HE" <[e-mail]>
>>>> Cc: <[e-mail]>; <[e-mail]>; <[e-mail]>; 
>>>> <[e-mail]>
>>>> Subject: RE: [mpls] [PWE3] 1588 over MPLS draft
>>>> 
>>>>> Hi Mach,
>>>>> 
>>>>> When a service provider wants to create these PTP LSPs, what is wrong 
>>>>> with allocating a range of labels for this purpose? This is purely a 
>>>>> software exercise. There are 2 million labels available to each node, 
>>>>> why can't some of them be allocated by software to PTP?
>>>>> 
>>>>> In theory what you say is correct and should work, but in practice 
>>>>> there is a function called 1-step Transparent clocking that requires 
>>>>> time stamping at the PHY (immediately when the packet is received or 
>>>>> transmitted). PHY chips don't have CAM or lots of memory to store a 
>>>>> few thousand random labels. The label range will solve that problem 
>>>>> and is consistent with MPLS architecture.
>>>>> 
>>>>> Thanks,
>>>>> Shahram
>>>>> 
>>>>> -----Original Message-----
>>>>> From: Mach Chen [mailto:[e-mail]]
>>>>> Sent: Wednesday, July 14, 2010 1:53 AM
>>>>> To: S. Davari; Jia HE; Shahram Davari
>>>>> Cc: [e-mail]; [e-mail]; [e-mail]; [e-mail]
>>>>> Subject: Re: [mpls] [PWE3] 1588 over MPLS draft
>>>>> 
>>>>> Hi Shahram,
>>>>> 
>>>>> From the view of implementation, there is no more difference between 
>>>>> SHOULD
>>>>> and MUST :-)
>>>>> For me, the Label Range is more like a mechanim to notify related 
>>>>> nodes that
>>>>> some LSPs are dedicated for PTP messages other than the chips memory
>>>>> limitation, because the memory restriction is always there whatever 
>>>>> you use
>>>>> Label Range or not.
>>>>> IMHO, since the objective is to tell related MPLS nodes which LSPs are 
>>>>> PTP
>>>>> LSPs, an indication in the signaling(RSVP-TE/GMPLS/LDP) is enough and 
>>>>> seems
>>>>> more common. And it will avoid the strict requirement of "the network and
>>>>> all nodes required to support the Label range".
>>>>> In addition, there should be some mechanims(e.g., ISIS/OSPF 
>>>>> extensions) for
>>>>> nodes to advertise their PTP capability hence to help PTP LSPs 
>>>>> computation.
>>>>> 
>>>>> Best regards,
>>>>> Mach
>>>>> 
>>>>> 
>>>>> --------------------------------------------------
>>>>> From: "S. Davari" <[e-mail]>
>>>>> Sent: Wednesday, July 14, 2010 2:31 PM
>>>>> To: "Jia HE" <[e-mail]>; "Shahram Davari" <[e-mail]>
>>>>> Cc: <[e-mail]>; <[e-mail]>; <[e-mail]>; 
>>>>> <[e-mail]>
>>>>> Subject: Re: [mpls] [PWE3] 1588 over MPLS draft
>>>>> 
>>>>>> Hi Jia,
>>>>>> 
>>>>>> Label range is a SHOULD requirements and not MUST. The reason for Label
>>>>>> Range is
>>>>>> mainly for PHY chips that don't have large memory and can't store a 
>>>>>> lot of
>>>>>> Labels. Otherwise the PTP LSP is setup via signaling that specifies the
>>>>>> LSP as
>>>>>> carrying 1588.
>>>>>> 
>>>>>> So the answer is that if a Label range is used it must be a Global range
>>>>>> within
>>>>>> a network and should not be used by any router for other applications.
>>>>>> 
>>>>>> Thanks,
>>>>>> Shahram
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> ________________________________
>>>>>> From: Jia HE <[e-mail]>
>>>>>> To: Shahram Davari <[e-mail]>
>>>>>> Cc: [e-mail]; [e-mail]; [e-mail]; [e-mail]
>>>>>> Sent: Tue, July 13, 2010 9:50:37 PM
>>>>>> Subject: Re: [mpls] [PWE3] 1588 over MPLS draft
>>>>>> 
>>>>>> 
>>>>>> Hi Shahram,
>>>>>> 
>>>>>> One question about "PTP Label Range":
>>>>>> 
>>>>>> To my knowledge, label in MPLS network is a local  matter. For 
>>>>>> example, we
>>>>>> may
>>>>>> have per-interface or per-platform label space. Will  this specificed 
>>>>>> "PTP
>>>>>> Label
>>>>>> Range" conflict with the current in-use labels  for common LSPs?
>>>>>> 
>>>>>> 
>>>>>> B.R.
>>>>>> Jia
>>>>>> 
>>>>>> ----- Original Message -----
>>>>>>> From: Shahram    Davari
>>>>>>> To: [e-mail] ; [e-mail] ; [e-mail] ; [e-mail]
>>>>>>> Sent: Thursday, July 08, 2010 3:12 AM
>>>>>>> Subject: [PWE3] 1588 over MPLS draft
>>>>>>> 
>>>>>>> 
>>>>>>> Hi All,
>>>>>>> 
>>>>>>> Please find attached our first draft of 1588 over MPLS.    Since we 
>>>>>>> have
>>>>>>> some
>>>>>>> technical issues converting the Word format to Txt we    couldn’t 
>>>>>>> upload
>>>>>>> the
>>>>>>> draft before the cut-off date. However we will    present the draft 
>>>>>>> in the
>>>>>>> next
>>>>>>> IETF meeting and will upload the draft after the    meeting.
>>>>>>> 
>>>>>>> Note that the main WG is TicToc but may require    consultation with 
>>>>>>> MPLS
>>>>>>> and
>>>>>>> PWE3 WGs.
>>>>>>> 
>>>>>>> Thanks,
>>>>>>> Shahram Davari
>>>>>> ________________________________
>>>>>> _______________________________________________
>>>>>>> pwe3 mailing    list
>>>>>>> [e-mail]
>>>>>>> https://www.ietf.org/mailman/listinfo/pwe3
>>>>>>> 
>>>>> 
>>>>>> _______________________________________________
>>>>>> mpls mailing list
>>>>>> [e-mail]
>>>>>> https://www.ietf.org/mailman/listinfo/mpls
>>>>>> 
>>>>> 
>>>> _______________________________________________
>>>> mpls mailing list
>>>> [e-mail]
>>>> https://www.ietf.org/mailman/listinfo/mpls
>> 
> 
> _______________________________________________
> mpls-tp mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/mpls-tp