Re: New draft about In-network Computing Solution

[Hesham ElBakoury <[e-mail]>]

I worked with professor Stefano on Extensible In-band Processing (EIP)
which uses a field in the packet to contain data and operations to be
performed by the network devices on this data. Data can be provided by the
user or collected using in-band telemetry or other means. EIP can be used
to support the use cases in your draft and others such as distributed
in-band machine learning, semantic routing, detnet, .... etc.

More info on EIP including a draft which describes EIP structure is
available in this page: https://eip-home.github.io/eip/

Thanks
Hesham

On Thu, Apr 27, 2023, 6:36 AM Zhe Lou <zhe.lou=[e-mail]>
wrote:

> Dear Kehan,
>
>
>
> Many thanks for your comments. I apologize for this late response, as I
> took a couple of weeks’ vacation. Please see my comments inline.
>
>
>
> Regards
>
> David
>
>
>
> *From:* 姚柯翰 <[e-mail]>
> *Sent:* Wednesday, 12 April 2023 06:16
> *To:* Zhangcuimin <[e-mail]>; Luigi IANNONE <
> [e-mail]>; zhouyujing (A) <[e-mail]>; Zhe
> Lou <[e-mail]>; yangjinze <[e-mail]>
> *Cc:* [e-mail]; [e-mail]
> *Subject:* Re:[sfc] New draft about In-network Computing
> Solution(draft-zhou-sfc-sinc)
>
>
>
> Dear authors,
>
>
>
> It's an interesting idea to consider about routing packets to specific
> network nodes for computation offloading. After reading the draft, I have
> the following comments:
>
>
>
> 1.     In-network computing(INC) is closely coupled with applications.
> It's about implementing application-specific functions inside commercial
> network devices. In section 5, the draft says  "In the SINC domain, a host
> MUST be SINC-aware. It defines the data operation to be executed.", I agree
> with the first sentence, but I think it might not be the case that data
> operation should be defined by applications, instead, I think these generic
> computing operations should be pre-defined by network and be open to
> applications through specific APIs.
>
> [DL] I think we are talking about 2 things, data calculation/operation
> requested by the application, and data operation capabilities offered by
> the network node. In the draft, we refer to the former. We do agree that
> the later is also important, as we briefly mentioned it in the “control
> plane requirements” section.
>
> To realize successful computation, apart from the definition of computing
> operations, other related terms should be defined as well, like data types,
> data structures and calculation precision, etc.  Since the capabilities of
> network devices provided by different vendors varies a lot, only
> tranferring computing operations might not be sufficient to perform a
> successful computation. If these definition can be done by network and
> encapsulated through common APIs which could be open to applications, and
> It's more friendly for App developers.
>
> [DL] The same as above, the application tell the network what it wants and
> the network tells the application what it can.
>
> 2.     The control plane requirements in section 8 are a little bit too
> general, even though specific control plane design is not in the scope of
> this draft. For example, in bullet one, what kind of resources in
> switches/routers should be awared? In bullet two, what kind of constraints
> should be based on for the selection of proper switches/routers to perform
> INC? These requirements should be more clear, and it might influence the
> control plane design.
>
> [DL] These are good points. We are going to update the section according
> to the comments from you and Jeff (from the IETF 116 meeting). If you have
> suggestions, we may work together on this part.
>
> 3.     The computing operations defined in Appendix A are somewhat
> atomic, and it feels like these operations are not enough for
> switches/routers to perform successful task offloading. IMHO, as I
> mentioned in comment 1, INC primitives should be presented through
> encapsulated APIs that network can open to applications. This might be more
> realistic.
>
> [DL] According to the discussion in IETF 116, the appendix A will be
> removed.
>
>
>
> Welcome more discussions and I'm very happy to contribute if needed.
>
>
>
> Best,
>
> Kehan
>
>
>
>
>
>
>
> ----邮件原文----
> 发件人："zhouyujing \\(A\\)" <zhouyujing3=[e-mail]>
> 收件人："[e-mail]" <[e-mail]>,"[e-mail]" <[e-mail]>
> 抄 送: Zhe Lou  <[e-mail]>
> 发送时间：2022-10-28 16:36:57
> 主题：
> [sfc] New draft about  In-network Computing Solution(draft-zhou-sfc-sinc)
>
> Hi all,
>
> We submitted a new draft about In-Network Computing (
> https://datatracker.ietf.org/doc/draft-zhou-sfc-sinc/
> ). In this draft, we want to discuss a mechanism "Signaling In-Network Computing operations" (SINC) to enable in-packet operation signaling for in-network computing for specific scenarios.
>
> Hope to get your review and comments.
>
> Many thanks!
>
> Best,
> Yujing Zhou
>
> -----Original Message-----
> From: [e-mail] <[e-mail]>
> Sent: 2022年10月24日 9:56
> To: Zhangcuimin <[e-mail]>; Luigi IANNONE <
> [e-mail]>; zhouyujing (A) <[e-mail]
> >; Zhangcuimin <[e-mail]>; Zhe Lou <[e-mail]>
> Subject: New Version Notification for draft-zhou-sfc-sinc-00.txt
>
>
>
> A new version of I-D, draft-zhou-sfc-sinc-00.txt has been successfully submitted by Yujing Zhou and posted to the IETF repository.
>
> Name: draft-zhou-sfc-sinc
> Revision: 00
> Title: Signaling In-Network Computing operations (SINC)
> Document date: 2022-10-23
> Group: Individual Submission
> Pages: 19
> URL:            https://www.ietf.org/archive/id/draft-zhou-sfc-sinc-00.txt
> Status:         https://datatracker.ietf.org/doc/draft-zhou-sfc-sinc/
> Html:
> https://www.ietf.org/archive/id/draft-zhou-sfc-sinc-00.html
> Htmlized:       https://datatracker.ietf.org/doc/html/draft-zhou-sfc-sinc
>
>
> Abstract:
>    This memo introduces "Signaling In-Network Computing operations"
>    (SINC), a mechanism to enable in-packet operation signaling for in-
>    network computing for specific scenarios like NetReduce,
>    NetDistributedLock, NetSequencer, etc.  In particular, this solution
>    allows to flexibly communicate computation parameters to be used in
>    conjunction with the packets' payload, to signal to in-network SINC-
>    enabled devices the computing operations to be performed.
>
>
>
>
>
> The IETF Secretariat
>
>
>
> _______________________________________________
> rtgwg mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/rtgwg
>
> Subject：
> [sfc] New draft about  In-network Computing Solution(draft-zhou-sfc-sinc)
>
> Hi all,
>
> We submitted a new draft about In-Network Computing (
> https://datatracker.ietf.org/doc/draft-zhou-sfc-sinc/
> ). In this draft, we want to discuss a mechanism "Signaling In-Network Computing operations" (SINC) to enable in-packet operation signaling for in-network computing for specific scenarios.
>
> Hope to get your review and comments.
>
> Many thanks!
>
> Best,
> Yujing Zhou
>
> -----Original Message-----
> From: [e-mail] <[e-mail]>
> Sent: 2022年10月24日 9:56
> To: Zhangcuimin <[e-mail]>; Luigi IANNONE <
> [e-mail]>; zhouyujing (A) <[e-mail]
> >; Zhangcuimin <[e-mail]>; Zhe Lou <[e-mail]>
> Subject: New Version Notification for draft-zhou-sfc-sinc-00.txt
>
>
>
> A new version of I-D, draft-zhou-sfc-sinc-00.txt has been successfully submitted by Yujing Zhou and posted to the IETF repository.
>
> Name: draft-zhou-sfc-sinc
> Revision: 00
> Title: Signaling In-Network Computing operations (SINC)
> Document date: 2022-10-23
> Group: Individual Submission
> Pages: 19
> URL:            https://www.ietf.org/archive/id/draft-zhou-sfc-sinc-00.txt
> Status:         https://datatracker.ietf.org/doc/draft-zhou-sfc-sinc/
> Html:
> https://www.ietf.org/archive/id/draft-zhou-sfc-sinc-00.html
> Htmlized:       https://datatracker.ietf.org/doc/html/draft-zhou-sfc-sinc
>
>
> Abstract:
>    This memo introduces "Signaling In-Network Computing operations"
>    (SINC), a mechanism to enable in-packet operation signaling for in-
>    network computing for specific scenarios like NetReduce,
>    NetDistributedLock, NetSequencer, etc.  In particular, this solution
>    allows to flexibly communicate computation parameters to be used in
>    conjunction with the packets' payload, to signal to in-network SINC-
>    enabled devices the computing operations to be performed.
>
>
>
>
>
> The IETF Secretariat
>
>
>
> _______________________________________________
> rtgwg mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/rtgwg
> _______________________________________________
> rtgwg mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/rtgwg
>