Re: [OPSAWG] Call for Adoption "draft-song-opsawg-ifit-framework"

[Haoyu Song <[e-mail]>]

Hi Frank,

IOAM is one of the techniques covered by iFIT, although an important one. In the draft, we include many others (we may omit some but we’ll include them once we are aware of them), the reason for that is also discussed in the draft: we don’t expect one technique can meet all the data plane telemetry requirements. In the future, it’s very likely we’ll end up with multiple options, each with its own applicability and advantages.

Also, this draft doesn’t meant to summarize such techniques. Rather, it provides a solution framework which may apply one or more such underlying techniques. While we have listed all the issues and challenges we know so far, we do want to add more operational experiences as the draft progresses. To this end, we request collaborations and contributions from all the operators and vendors. We believe this will benefit the entire community.

Best,
Haoyu

From: Frank Brockners (fbrockne) <[e-mail]>
Sent: Friday, October 25, 2019 7:03 AM
To: qinfengwei <[e-mail]>; Haoyu Song <[e-mail]>; Joe Clarke (jclarke) <[e-mail]>
Cc: [e-mail]; [e-mail]
Subject: RE: [OPSAWG] Call for Adoption "draft-song-opsawg-ifit-framework"

Hi Fengwei Qin,

thanks a lot for the additional context. You confirm that “iFIT nodes” etc. are indeed new functions with data plane and control plane functionality that refer to a specific deployment that you’ve done. BTW/ - as a side note: It was surprising to me that despite your deployment seems to be based on draft-cheng-mpls-inband-pm-encapsulation – the draft-song-opsawg-ifit-framework does not even mention it.
We’re all aware that there are several proprietary solutions that resemble IOAM.  We started the IOAM effort several years ago in IETF to jointly arrive at a standard that allows for multivendor interoperable solutions.  Focusing on yet another variant of IOAM isn’t really moving the industry forward.

Rather than trying to write a generic document which seems to try to provide an overview of “what is out there” it would be much more helpful, if you could provide a detailed description of the issues and experiences (more specific and detailed than what you have in section 1 of draft-song-opsawg-ifit-framework) that you encountered in your deployment. Based on that understanding, we could jointly explore how to evolve the existing standardization effort on IOAM.

Thanks, Frank

From: qinfengwei <[e-mail]<mailto:[e-mail]>>
Sent: Freitag, 25. Oktober 2019 07:35
To: Frank Brockners (fbrockne) <[e-mail]<mailto:[e-mail]>>; 'Haoyu Song' <[e-mail]<mailto:[e-mail]>>; Joe Clarke (jclarke) <[e-mail]<mailto:[e-mail]>>
Cc: [e-mail]<mailto:[e-mail]>; [e-mail]<mailto:[e-mail]>
Subject: 答复: [OPSAWG] Call for Adoption "draft-song-opsawg-ifit-framework"

Hi Frank,

Speaking as the co-author, I think the abstraction described well the scope:

   In-situ Flow Information Telemetry (iFIT) provides a method for
   efficiently applying underlying on-path flow telemetry techniques,
   applicable across various network environments.

   This document outlines the iFIT framework, which enumerates several
   critical functional components and describes how these components are
   assembled together to achieve a complete and closed-loop working
   solution for on-path flow telemetry.

We, China Mobile, have deployed one IFIT data plane with many vendor supports.
https://datatracker.ietf.org/doc/draft-cheng-mpls-inband-pm-encapsulation/<https://nam03.safelinks.protection.outlook.com/?url=https%3A%2F%2Fdatatracker.ietf.org%2Fdoc%2Fdraft-cheng-mpls-inband-pm-encapsulation%2F&data=02%7C01%7Chaoyu.song%40futurewei.com%7C1138aa61827241949d1a08d75954129d%7C0fee8ff2a3b240189c753a1d5591fedc%7C1%7C0%7C637076089965871536&sdata=IrxMLx0vMPvvXoOuRTdly2C0V%2FndlZDzzM8YP%2FKQFeE%3D&reserved=0>

We compared and chose the suitable data plane. We considered the forwarding performance.
We also considered how to reduce the data export to the controller. And so on.
Data plane is not the only thing as in a complete solution.
We want to see this be generalized in a framework to achieve a complete and closed-loop working solution for on-path flow telemetry.

The document is not complete yet, some parts like section 3.6 and 4 are newly added into the draft according to the comments from the mailing list.
Your contributions are welcome.

Thanks,
Fengwei Qin

发件人: OPSAWG [mailto:[e-mail]] 代表 Frank Brockners (fbrockne)
发送时间: 2019年10月25日 00:42
收件人: Haoyu Song; Joe Clarke (jclarke)
抄送: [e-mail]<mailto:[e-mail]>; [e-mail]<mailto:[e-mail]>
主题: Re: [OPSAWG] Call for Adoption "draft-song-opsawg-ifit-framework"

Just took a read through the document as well – and I can echo Joe’s comments.
The scope of the document is not clear, nor does one understand what problem iFIT would address and solve.
That said, the document seems to have a very specific implementation in mind, as it refers to specific things such “iFIT Applications”, “iFIT Nodes”, etc. – but none of things are defined in the document.

Cheers, Frank

From: OPSAWG <[e-mail]<mailto:[e-mail]>> On Behalf Of Haoyu Song
Sent: Mittwoch, 23. Oktober 2019 11:43
To: Joe Clarke (jclarke) <[e-mail]<mailto:[e-mail]>>
Cc: [e-mail]<mailto:[e-mail]>; [e-mail]<mailto:[e-mail]>
Subject: Re: [OPSAWG] Call for Adoption "draft-song-opsawg-ifit-framework"

Hi Joe,

Thank you for the detailed comments!

Let me try to explain the purpose of this draft better. Given the recent on-path data plane telemetry techniques and standard works, in this draft we discuss the deployment challenges and potential opportunities for applications. There is no such a document in IETF AFAIK and we feel it’s needed (also confirmed by some network operators who are interested in such techniques)

Most related standard proposals so far only defines the data plane protocol and lack considerations for a complete solution. To this end, we discuss various points that a solution should pay attention to and how these can be composed to support applications. Along with the discussion, we provide some examples and use cases to trigger new ideas.

We deliberately make iFIT an open framework and avoid introducing any new protocol and enforcing any specific approaches, because otherwise we are in danger to put unnecessary constraints on implementation approaches and hurt the possibility of innovation. While we mean to keep this document informational, we may consider to add more discussions on reference designs, operational experiences,  and best practices as you suggested.

Some points you raised below also deserves more detailed explanation, such as how to make an iFIT closed loop and how architecture and algorithm components can be composed to form such a loop. Perhaps a complete example can help to explain that. I’ll consider all this in future revisions.

In a sense, this document indeed aims to discuss the implementation, operational experiences, and best practices  of PBT, IOAM, and other similar techniques. We hope this document will trigger new drafts on management plane/control plane and innovative solutions.

Best regards,
Haoyu


From: Joe Clarke (jclarke) <[e-mail]<mailto:[e-mail]>>
Sent: Tuesday, October 22, 2019 3:48 PM
To: Haoyu Song <[e-mail]<mailto:[e-mail]>>
Cc: [e-mail]<mailto:[e-mail]>; [e-mail]<mailto:[e-mail]>
Subject: Re: Call for Adoption "draft-song-opsawg-ifit-framework"

The comments below are from me as an individual contributor.

I have read the latest revision of this document.  I still do not have a clear idea of what it is solving, TBH.  It doesn’t define a new protocol, yet it makes claims about an architecture that implies a protocol between devices, a controller, and applications (see the figure in section 2).  In Section 3.2, iFIT is referenced as having an ability to cache or send accumulated data.  I don’t see how a framework can do this.  Nor do I see how a framework can dynamically load new data probes as mentioned in Section 3.3.  If this were a controller with an application architecture and specifications for component interoperability, perhaps, but I do not see that in this document.

In Section4, the document mentions a closed-loop for iFIT applications whereby applications can manage iFIT closed loops on top of a controller.  But again, I don’t see how.  Do the applications make API calls?  What calls do they make?  What makes it an “iFIT closed loop”?

Ultimately, the summary says that iFIT combines algorithmic and architectural schemes into the framework, but I don’t see where that is done in a specific, implementable way (e.g., in Section 3.1.2 you begin to describe how you can adaptively sample packets, but you talk about abstract signals to/from the controller).  Nor do I see how one would implement iFIT.  When I read the iFIT draft, I feel like I’m missing a normative chunk that explains how the various pieces of this framework are to interact in a well-specified manner.

It seems to me that perhaps a more useful document is one that focus on the implementation of PBT and/or IOAM, operational experiences, best practices, etc.

Joe

On Oct 21, 2019, at 14:34, Haoyu Song <[e-mail]<mailto:[e-mail]>> wrote:

Dear OPSAWG chairs,

The following draft has been extensively discussed and gone through six revisions. Network operators confirmed it is useful.
We believe the draft is mature enough to be adopted by the WG therefore we request the chairs to initiate the adoption call for this draft.
Thank you very much for the consideration!

https://datatracker.ietf.org/doc/draft-song-opsawg-ifit-framework/<https://nam03.safelinks.protection.outlook.com/?url=https%3A%2F%2Fdatatracker.ietf.org%2Fdoc%2Fdraft-song-opsawg-ifit-framework%2F&data=02%7C01%7Chaoyu.song%40futurewei.com%7C1138aa61827241949d1a08d75954129d%7C0fee8ff2a3b240189c753a1d5591fedc%7C1%7C0%7C637076089965881506&sdata=dk6jjUHDXzRWMdXp3dPUn1U4DvTHY5I%2Bdj3YzwZwYF8%3D&reserved=0>

Best regards,
Haoyu