<?xml version="1.0" encoding="US-ASCII"?>
<!-- This is built from a template for a generic Internet Draft. Suggestions for
     improvement welcome - write to Brian Carpenter, brian.e.carpenter @ gmail.com 
     This can be converted using the Web service at http://xml.resource.org/ -->
<!DOCTYPE rfc SYSTEM "rfc2629.dtd">
<!-- You want a table of contents -->
<!-- Use symbolic labels for references -->
<!-- This sorts the references -->
<!-- Change to "yes" if someone has disclosed IPR for the draft -->
<!-- This defines the specific filename and version number of your draft (and inserts the appropriate IETF boilerplate -->
<?rfc sortrefs="yes"?>
<?rfc toc="yes"?>
<?rfc symrefs="yes"?>
<?rfc compact="yes"?>
<?rfc subcompact="no"?>
<?rfc topblock="yes"?>
<?rfc comments="no"?>
<rfc category="info" docName="draft-liu-coinrg-requirement-03"
     ipr="trust200902">
  <front>
    <title abbrev="Computing in network">Requirement of Computing in
    network</title>

    <author fullname="Peng Liu" initials="P." surname="Liu">
      <organization>China Mobile</organization>

      <address>
        <postal>
          <street/>

          <city>Beijing</city>

          <code>100053</code>

          <country>China</country>
        </postal>

        <email>liupengyjy@chinamobile.com</email>
      </address>
    </author>

    <author fullname="Liang Geng" initials="L." surname="Geng">
      <organization>China Mobile</organization>

      <address>
        <postal>
          <street/>

          <city>Beijing</city>

          <code>100053</code>

          <country>China</country>
        </postal>

        <email>gengliang@chinamobile.com</email>
      </address>
    </author>

    <!---->

    <date day="12" month="July" year="2020"/>

    <area>Computing in Network</area>

    <workgroup>Computing in Network Research Group</workgroup>

    <keyword>Computing, Network, Requirement</keyword>

    <abstract>
      <t>New technology such as IOT, edge computing,etc. propose the
      requirement of computing in network, so the convergence of network and
      computing has become a trend. It will bring some new directions and
      areas to be considered, such as the relationship between network and
      computing, the influence of integrating computing to the network, and so
      on.</t>

      <t>This document points out the requirements of computing in network
      according to the development of new Industry, including the network and
      computing requirements.</t>
    </abstract>

    <note title="Requirements Language">
      <t>The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
      "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
      document are to be interpreted as described in <xref
      target="RFC2119">RFC 2119</xref>.</t>
    </note>
  </front>

  <middle>
    <section anchor="intro" title="Overview">
      <t>The new services' provider expects a user experience with lower
      latency and high reliability, which put forwards immense challenges to
      cloud computing and traditional network. Centralized computing requires
      a long transmission distance of traffic flow, and the existing network
      technology is to the best of its ability. Network operators start to
      think about how to meet the higher needs of service provider and users.
      Computing in the network may solve the problems because it can provide a
      flexible network and computing integration system.</t>

      <t>To integrate the computing resource to the network, it need to find
      suitable computing nodes to handle service's request, as well as a
      forwarding path to them. How much computing resources will affect the
      delay of service processing, which could also affect the whole network
      latency. Just as the measurement of network performance has one more
      dimension, it will interact and cooperate with others. So there are some
      requirments for both network and computing.</t>
    </section>

    <section title="Requirements of Network">
      <t>The network requirements includes precision, concurrent, addressing
      and information interaction.</t>

      <section title="Precision ">
        <t>Precision of the network refers to the deterministic of latency,
        packet loss rate and perception of computing resources.</t>

        <t>* Latency: The traditional network's best-effort forwarding mode
        can no longer meet the demand of such services for network latency.
        The deterministic latency brings forward a new measure latitude for
        network, which changes from in-time to on-time.</t>

        <t>* Packet loss rate: It is another factor to evaluate the precision
        of the network. Utilizing the ubiquitous computing capability of the
        network, network prediction and segment-by-segment path retransmitting
        are realized based on AI, network transmission can be optimized and
        service QoS can be ensured.</t>

        <t>* Perception of computing resources: how to precisely obtain the
        status of computing resource to meet the requirements of business
        requests is also a challenge to the network. It considers the network
        status and the performance status of computing resources can be
        matched dynamicly. So the user experience, utilization rate of
        computing resources and the network efficiency can be optimum.</t>

        <t>For the latency and packet loss rate, some technologies such as
        time-sensitive network TSN, deterministic network DetNet, etc., have
        proposed corresponding technical means to provide network bearers with
        deterministic latency(IEEE802.1Qbv, IEEE802.1Qbu) and packet loss rate
        and guarantee the user's business experience. However, it also needs
        to consider how to guarantee the service's end-to-end latency, packet
        loss rate and resource utilization rate.</t>

        <t>For the perception of computing resources, we may consider about
        the OAM and telemetry to achieve it, however, the performance and
        information collection strategies are issues that need attention.</t>
      </section>

      <section title="Concurrent">
        <t>There will be number of computing nodes deployed in the network, or
        computing fucntions intergrated in the network device for network
        computing. A serivce's computing request may distributed in several
        computing nodes in order to response quickly to the client. So there
        may be a lot of parallel computing task, whitch cause too much
        connection among the nodes but consume less bandwidth. It will bring
        great challenges to the concurrent network connection including how to
        build and deploy these distributed computing nodes to ensure the
        processing capability of the network , as well as the storage, call of
        the database are worth studying.</t>
      </section>

      <section title="Addressing">
        <t>Traditional application-based addressing can not accurately grasp
        the network performance in real time. The comprehensive performance of
        addressing results based on application layer may not be the best. It
        is always to find the consistent host's address and go through a long
        distance internet, which results in poor business experience.</t>

        <t>It need to find some new way to improve the addressing proccess.
        For example, in the function based addressing, the application
        deconstruction components on the server side are distributed on the
        cloud platform, and the business logic in the server is transferred to
        the client side. So the client only needs to care about the computing
        function itself, not about the computing resources such as server,
        virtual machine, container and so on.</t>
      </section>

      <section title="Information interaction">
        <t>The network needs to have the ability to sense application's
        requirments and expose network and computing status. For example,
        application can tell the network requirements including bandwidth,
        latency and jitter, as well as the computing requirements, such as
        CPU, storage and memory. The network also can have the capability to
        be aware of the application's requirements. Thus it can effectively
        support the network programming, which could meet the future business
        requirements.</t>
      </section>
    </section>

    <section title="Requirements of computing">
      <t>The computing requirements includes computing resource deployment,
      discovery, reservation, scheduling and OAM.</t>

      <section title="Computing resource deployment">
        <t>If some computing tasks in the network is planned to be
        implemented, it needs to consider about what kinds of chips and where
        should them be deployed. On the one hands, different kinds of
        computing require different kinds of chips, such as CPU, GPU and
        memory chip. On the other hands, those chips may be put into router,
        switch, server or some dedicated machines, which are connected by the
        network.</t>

        <t>There is an example about AI algorithm which might be discussed
        before. The AI algorithm has several steps including training and
        matching, and they also have different requirements of chips. In
        network computing, those steps could be distributed in different
        computing nodes.</t>
      </section>

      <section title="Computing resource discovery">
        <t>The network needs to have the ability to discover computing
        resources. when the computing nodes are deployed in the network, it
        need to be registered to the network management system, and the
        information of computing resource or routing can update. In this way,
        when there are computing tasks to be executed, the network can
        reasonably allocate resources according to the needs of the
        application.</t>
      </section>

      <section title="Computing resource reservation">
        <t>There might be serial distributed computing model of computing in
        the network, and different resources need to be reserved for different
        nodes. For example, AI algorithm now has a model of step-by-step
        iteration at multiple nodes. The previous iteration will affect the
        next calculation results, and the computing resources required for
        each iteration are not the same. From the perspective of network
        standard, we hope to regard computing resources as the dimensions to
        measure network performance, such as the same bandwidth, path, etc.,
        while the traditional technologies of resource reservation have not
        considered the reservation of computing resources, and have not
        considered the differentiated resource reservation model. Therefore,
        new protocol or extension of existing protocol is needed to meet the
        requirement.</t>
      </section>

      <section title="Computing aware scheduling">
        <t>Computing in network needs a reasonable scheduling strategy, which
        means computing aware scheduling. According to the business requests,
        dynamicly computing power matching is carried out based on network
        status and performance of computing resources to achieve optimal user
        experience, optimal utilization of computing resources and optimal
        network efficiency. In computing aware scheduling, computing is seen
        as &ldquo;link state&rdquo; and the computing resource information
        should be exposed.</t>
      </section>

      <section title="Computing resource OAM">
        <t>The ability of OAM can be used to continuously update the current
        computing power resources, and perform some troubleshooting tasks.
        However, OAM of computing resource is more complex than network.
        Network monitoring is relatively simple, like bandwidth, latency,
        jitter, while computing can be divided into many categories, different
        application need different kinds of computing. So it need to implement
        fine-grained OAM of computing resource.</t>
      </section>
    </section>

    <section title="Requirements of management">
      <t>The management requirements includes cross domain management, joint
      optimisation and multi user access.</t>

      <section title="Cross domain management">
        <t>The computing in the network should ensure the end-to-end network
        management to meet the needs of different network topology,
        performance and function, which involves cross domain network
        arrangement. In the process of network data transmission, different
        services will forward in different ways or different network
        protocols, and computing resources may be distributed in different
        network domains. Effective cross domain management will enhance the
        performance of network and computing.</t>
      </section>

      <section title="Joint optimisation">
        <t>As computing resources are integrated in the network, and may be
        used as one of the measurement dimensions of network performance,
        joint optimization is also a very important part. Network and
        computing resources will affect each other, including performance,
        scheduling and so on. So It need a good joint optimization scheduling
        strategy.</t>
      </section>

      <section title="Multi user access">
        <t>Many existing applications, such as games, remote video
        conferencing, are usually multi--accessed and interacted by several
        users at the same time. This brings about the problem of service
        consistency, that is, users accessing to the same game or video need
        the consistency of SLA, otherwise it will seriously affect the
        experience of other users. Service consistency can be achieved through
        network management or application layer control.</t>
      </section>
    </section>

    <section title="Conclusion">
      <t>Based on the requirements of new business, this document puts forward
      the requirements of computing in network, and gives some reference
      technologies and use cases. Computing in network is a new direction,
      some details need more in-depth discussion and research.</t>
    </section>

    <section anchor="Security" title="Security Considerations">
      <t>Computing In network has brought the trend of network convergence in
      different regions. For example, 5g network of operators can go deep into
      the vertical industry user site to provide users with higher quality
      network services, which will bring the convergence of operator's network
      and user site network. Besides, industrial Internet brings the trend of
      integration of industrial OT network and IT network to further improve
      the production efficiency of the industry. It need to ensure the
      security of the network, including the mutual trust and non aggression
      of information among regions, which may require further protection and
      detection measures.</t>
    </section>

    <section anchor="IANA" title="IANA Considerations">
      <t>TBD.</t>
    </section>
  </middle>

  <back>
    <references title="Normative References">
      <?rfc include="reference.RFC.2119"?>

      <?rfc include="reference.I-D.kutscher-coinrg-dir"?>
    </references>
  </back>
</rfc>
