<?xml version="1.0" encoding="US-ASCII"?>
<!DOCTYPE rfc SYSTEM "rfc2629.dtd">
<?rfc toc="yes"?>
<?rfc tocompact="yes"?>
<?rfc tocdepth="3"?>
<?rfc tocindent="yes"?>
<?rfc symrefs="yes"?>
<?rfc sortrefs="yes"?>
<?rfc comments="yes"?>
<?rfc inline="yes"?>
<?rfc compact="yes"?>
<?rfc subcompact="no"?>
<rfc category="info" docName="draft-fu-coinrg-joint-optimization-req-00"
     ipr="trust200902">
  <front>
    <title abbrev="Data-driven Accounting in APN6">Requirements of computing
    and network joint optimization and scheduling</title>

    <author fullname="Yuexia Fu" initials="Y." surname="Fu">
      <organization>China Mobile</organization>

      <address>
        <postal>
          <street>No.32 XuanWuMen West Street</street>

          <city>Beijing</city>

          <code>100053</code>

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

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

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

      <address>
        <postal>
          <street>No.32 XuanWuMen West Street</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>No.32 XuanWuMen West Street</street>

          <city>Beijing</city>

          <code>100053</code>

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

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

    <date month="November" year="2020"/>

    <area/>

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

    <keyword>computing, joint optimization</keyword>

    <abstract>
      <t>With the development of edge computing, there is a trend that
      computing is widely deployed in network rather than at other end of
      network, and provides services at nearer location. With the deep
      integration of network, traditional optimization and scheduling within
      network domain is not enough, the endpoint of the path matters a lot. So
      the relationship between computing and network are new and important
      topics to be studied. This document focus on the requirements of
      computing and network joint optimization and scheduling based on the
      newly arising service 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 .</t>
    </note>
  </front>

  <middle>
    <section title="Overview">
      <t>For traditional services without strict service requirements, the
      best-effort network can meet the requirements with traditional path
      optimization, which only consider the network condition. With new
      services arising, such as cloud AR/VR, cloud gaming, V2X, new and strict
      requirements towards network, also towards the service endpoint are
      proposed to meet the service requirements. So the computing and network
      joint optimization and scheduling are proposed to guarantee the service
      performance.</t>

      <t>The computing and network joint optimization means that there is not
      only the path optimization in network, but also the endpoint joint
      optimization; also the two will affect each other. Based on the joint
      optimization, the service scheduling can be performed considering the
      network condition and also the endpoint condition, with the
      &ldquo;optimal path+ optimal endpoint&rdquo; policy. What&rsquo;s more,
      the computing and network resources joint reservation is required for
      services with strict performance requirements.</t>

      <t/>
    </section>

    <section title="Requirements of awareness of computing">
      <t>The service requirements arising include both network and computing
      requirements, which further require future network should perform the
      joint optimization according to service requirements. So the
      requirements towards the joint optimization are: the awareness of
      computing and network requirements, the awareness of computing resources
      and services in network, the computing-aware path optimization and the
      network-aware endpoint optimization.</t>

      <t>2.1. Awareness of computing and network requirements</t>

      <t>Awareness of computing and network requirements refers to consider
      the computing requirements in addition to the network requirements,
      including the awareness of computing requirements and the measurement of
      computing services.</t>

      <t>Since network requirements can be measured with bandwidth, delay etc,
      it is also required to measure computing requirements in a unified way.
      On the one hand, there are many different computing services which are
      the &ldquo;consumers&rdquo; of computing resources, such as video
      processing, image classification etc, and they propose various
      requirements towards computing. It is required to firstly obtain the
      computing requirements and then model the requirements in a unified way,
      which then can be used as the constraint of joint optimization.</t>

      <t>What&rsquo;s more, the computing service modes are abundant compared
      to network services, which are the computing &ldquo;producers&rdquo;,
      including there are heterogeneous hardware such as GPU, CPU, FPGA etc,
      and also various algorithms deployed in network, so it is also required
      to model the computing producers in a unified way, which is another
      important factor for joint optimization. As for the awareness of
      computing requirements, some technologies such as application-aware
      networking have proposed corresponding technical solutions to delivery
      computing requirements in the packet head, however, it needs further
      study on the security of application and also the efficiency of the
      information delivery. As for the measurement of computing services,
      there is no mature solution to model the computing requirements and the
      computing resources in a unified way, which is a challenge for the
      computing and network joint optimization.</t>

      <t>2.2. Awareness of computing resources and services</t>

      <t>With the development of edge computing, the computing resources and
      computing services will be distributed in network, since the limited
      physical conditions, each computing site will be small scale and with
      limited computing resources, so different from the cloud computing,
      which can finish the computing task within one site, the edge computing
      sites need the collaboration among many sites, and this collaboration
      can be done in network. To coordinate the computing sites, it is
      required for network to be aware of the computing status of edge sites,
      including the real-time status of computing resources and computing
      services. So how to generate the required information and then broadcast
      it to network brings new challenges.</t>
    </section>

    <section title="Requirements of computing and network joint optimization">
      <t>3.1 Computing-aware path optimization</t>

      <t/>

      <t>With new services requiring computing and network resources,
      traditional network-based path optimization can not accurately guarantee
      the service requirements. The network-based path optimization only
      according to network conditions can only make sure the performance of
      network services, it can only find a best path towards a given endpoint,
      however, the given endpoint may be not optimal, causing the service
      requirements cannot be met.</t>

      <t>So It is required to do the computing-aware path optimization to
      consider the status of endpoint. For example, before the path
      optimization, according to the awareness of computing resources and
      services in network, including the location and status, the network
      could firstly find a list of optimal computing nodes, then the network
      could do path optimization with different computing endpoints, which
      changes the traditional way to only do the path optimization with one
      destination.</t>

      <t/>

      <t>To better optimize the computing-aware path, we need to consider
      different weights of computing and network metrics when calculating the
      optimal path. For traditional path optimization, there are only network
      metrics as the parameters of algorithm; it is required to add computing
      metrics also as the calculation metrics of the algorithm and to combine
      the computing and network metrics.</t>

      <t>What&rsquo;s more, based on the awareness of service requirements,
      for different services, there will be different requirements towards
      computing and network. For some computing-intensive services, computing
      counts more on the whole process of services, so they will require more
      on computing than network; and for communication-intensive services, the
      computing is less during the service process, while there will be
      frequent communication, which will propose higher requirement towards
      network than computing. So it can be inferred that computing and network
      matters differently during the service process for various services.</t>

      <t>Based on what discussed above, it is required to adaptively define
      different weights of computing and network metrics for different
      services, adapting to various service requirements. For example, for the
      computing-intensive services, it is required to put more weights on
      computing metrics than network metrics, which could be based on the
      percentage of predicted computing time in whole time; as for
      communication-intensive services, more weights could be put to adapt to
      the service requirements.</t>

      <t>3.2 Network-aware endpoint optimization</t>

      <t/>

      <t>Based on the computing-aware path optimization, there will be the
      optimal &ldquo;path + endpoint&rdquo; pair, combing the computing and
      network status. But there will also be inner scheduling in computing
      node, which may also influence the computing time. With proper task
      assignment, the computing time could be less to make sure that endpoint
      provides the promised services. So it is also required for endpoint to
      know the service requirements precisely, otherwise the endpoint will
      just do the usual scheduling without considering the service
      requirements.</t>

      <t>With the network-awareness, the endpoint will know the performance of
      network, such as the endpoint will know the transmission time in network
      and then calculate the rest of required time, and then it will do the
      inner scheduling accordingly.</t>

      <t/>
    </section>

    <section title="Requirements of computing and network joint resource reservation">
      <t/>

      <t>For services with strict computing requirements, the resource
      reservation should include network reservation and computing
      reservation, also, the two will affect each other.</t>

      <t>There is network resource reservation in traditional QoS guarantee
      mechanism based on the network resources reservation calculation to
      reserve specific resources for specific services. With new services
      arising, the network resources reservation is not enough, since the
      completion of services include not only network transmission but also
      endpoint calculation, only reserving the network resources cannot make
      sure the required computing resources are available during the required
      time for specific service.</t>

      <t>So facing the trend of computing and network convergence, it is also
      required to reserve the computing resources together with the network
      resources. Based on the awareness of service requirements and the joint
      path optimization, it is required to map the computing requirements into
      the corresponding computing resources reservation, for example, to map
      the services type into the computing resources type, and translate the
      computing latency requirements towards the required amount computing
      resources.</t>

      <t>On the other hand, the reservation of network and computing resources
      are closely linked, there will be different network resource reservation
      policy considering the computing resources reservation. For example, the
      order of the two resources reservation requires to be considered since
      they are relative independent usually.</t>

      <t>What&rsquo;s more, it is also required to dynamically adjust the
      resources reservation according to real-time status. One scenario is
      that the computing resource reservation could be adjusted based on the
      information from network domain, including the reservation time and also
      the reservation amount. Another scenario is the co-adjust of the two
      resources reservation, in network domain, the path and the relative
      reservation could be adjusted, and then the computing domain is required
      to adjust on-demand.</t>
    </section>

    <section title="Conclusion ">
      <t>Based on the new services&rsquo; requirements on computing and
      network, this document puts forward requirements of computing and
      network joint optimization, and also proposes requirements of computing
      and network joint resource reservation. Computing in network is a new
      direction, how to collaborate computing and network need further
      study.</t>
    </section>

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

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

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

  <back>
    <references title="Informative References">
      <?rfc include='reference.I-D.li-apn6-problem-statement-usecases'?>
    </references>
  </back>
</rfc>
