Skip to main content

Considerations of deploying AI services in a distributed method
draft-hong-nmrg-ai-deploy-05

Document Type Active Internet-Draft (individual)
Authors Yong-Geun Hong , Oh Seokbeom , Joo-Sang Youn , SooJeong Lee , Seung-Woo Hong , Ho-Sun Yoon
Last updated 2023-10-23
RFC stream (None)
Intended RFC status (None)
Formats
Stream Stream state (No stream defined)
Consensus boilerplate Unknown
RFC Editor Note (None)
IESG IESG state I-D Exists
Telechat date (None)
Responsible AD (None)
Send notices to (None)
draft-hong-nmrg-ai-deploy-05
Internet Research Task Force                                   Y-G. Hong
Internet-Draft                                        Daejeon University
Intended status: Informational                                   S-B. Oh
Expires: 25 April 2024                                               KSA
                                                               J-S. Youn
                                                           DONG-EUI Univ
                                                                S-J. Lee
                                                     Korea University/KT
                                                               S-W. Hong
                                                               H-S. Yoon
                                                                    ETRI
                                                         23 October 2023

    Considerations of deploying AI services in a distributed method
                      draft-hong-nmrg-ai-deploy-05

Abstract

   As the development of AI technology matured and AI technology began
   to be applied in various fields, AI technology is changed from
   running only on very high-performance servers with small hardware,
   including microcontrollers, low-performance CPUs and AI chipsets.  In
   this document, we consider how to configure the network and the
   system in terms of AI inference service to provide AI service in a
   distributed method.  Also, we describe the points to be considered in
   the environment where a client connects to a cloud server and an edge
   device and requests an AI service.  Some use cases of deploying AI
   services in a distributed method such as self-driving car and digital
   twin network are described.

Status of This Memo

   This Internet-Draft is submitted in full conformance with the
   provisions of BCP 78 and BCP 79.

   Internet-Drafts are working documents of the Internet Engineering
   Task Force (IETF).  Note that other groups may also distribute
   working documents as Internet-Drafts.  The list of current Internet-
   Drafts is at https://datatracker.ietf.org/drafts/current/.

   Internet-Drafts are draft documents valid for a maximum of six months
   and may be updated, replaced, or obsoleted by other documents at any
   time.  It is inappropriate to use Internet-Drafts as reference
   material or to cite them other than as "work in progress."

   This Internet-Draft will expire on 25 April 2024.

Hong, et al.              Expires 25 April 2024                 [Page 1]
Internet-Draft            Deploying AI services             October 2023

Copyright Notice

   Copyright (c) 2023 IETF Trust and the persons identified as the
   document authors.  All rights reserved.

   This document is subject to BCP 78 and the IETF Trust's Legal
   Provisions Relating to IETF Documents (https://trustee.ietf.org/
   license-info) in effect on the date of publication of this document.
   Please review these documents carefully, as they describe your rights
   and restrictions with respect to this document.  Code Components
   extracted from this document must include Revised BSD License text as
   described in Section 4.e of the Trust Legal Provisions and are
   provided without warranty as described in the Revised BSD License.

Table of Contents

   1.  Introduction  . . . . . . . . . . . . . . . . . . . . . . . .   3
   2.  Procedure to provide AI services  . . . . . . . . . . . . . .   5
   3.  Network configuration structure to provide AI services  . . .   6
     3.1.  AI inference service on Local machine . . . . . . . . . .   6
     3.2.  AI inference service on Cloud server  . . . . . . . . . .   7
     3.3.  AI inference service on Edge device . . . . . . . . . . .   8
     3.4.  AI inference service on Cloud server and Edge device  . .   9
     3.5.  AI inference service on horizontal multiple servers . . .  10
     3.6.  Network-side utilization for AI learning  . . . . . . . .  11
   4.  Considerations for configuring a network to provide AI
           services  . . . . . . . . . . . . . . . . . . . . . . . .  12
     4.1.  Considerations according to the functional characteristics
           of the hardware . . . . . . . . . . . . . . . . . . . . .  12
     4.2.  Considerations according to the characteristics of the AI
           model . . . . . . . . . . . . . . . . . . . . . . . . . .  13
     4.3.  Considerations according to the characteristics of the
           communication method  . . . . . . . . . . . . . . . . . .  14
   5.  Use cases of deploying AI services in a distributed method  .  14
     5.1.  Deploying AI services in Self-driving car . . . . . . . .  15
     5.2.  Deploying AI services in Digital twin network . . . . . .  16
   6.  IANA Considerations . . . . . . . . . . . . . . . . . . . . .  19
   7.  Security Considerations . . . . . . . . . . . . . . . . . . .  19
   8.  Acknowledgements  . . . . . . . . . . . . . . . . . . . . . .  19
   9.  Informative References  . . . . . . . . . . . . . . . . . . .  19
   Authors' Addresses  . . . . . . . . . . . . . . . . . . . . . . .  20

Hong, et al.              Expires 25 April 2024                 [Page 2]
Internet-Draft            Deploying AI services             October 2023

1.  Introduction

   In the Internet of Things (IoT), the amount of data generated from
   IoT devices has exploded along with the number of IoT devices due to
   industrial digitization and the development and dissemination of new
   devices.  Various methods are being tried to effectively process the
   explosively increasing IoT devices and data of IoT devices.  One of
   them is to provide IoT services in a place located close to IoT
   devices and users, away from cloud computing that transmits all data
   generated from IoT devices to a cloud server
   [I-D.irtf-t2trg-iot-edge].

   IoT services also started to break away from the traditional method
   of analyzing IoT data collected so far in the cloud and delivering
   the analyzed results back to IoT objects or devices.  In other words,
   AIoT (Artificial Intelligence of Things) technology, a combination of
   IoT technology and artificial intelligence (AI) technology, started
   to be discussed at international standardization organizations such
   as ITU-T.  AIoT technology, discussed by the ITU-T CG-AIoT group, is
   defined as a technology that combines AI technology and IoT
   infrastructure to achieve more efficient IoT operations, improve
   human-machine interaction, and improve data management and analysis
   [CG-AIoT].

   The first work started by the IETF to apply IoT technology to the
   Internet was to research a lightweight protocol stack instead of the
   existing TCP/IP protocol stack so that various types of IoT devices,
   not traditional Internet terminals, could access the Internet
   [RFC6574][RFC7452].  These technologies have been developed by
   6LoWPAN working group, 6lo working group, 6tisch working group, core
   working group, t2trg group, etc.  As the development of AI technology
   matured and AI technology began to be applied in various fields, just
   as IoT technology was mounted on resource-constrained devices and
   connected to the Internet, AI technology is also changed from running
   only on very high-performance servers.  The technology is being
   developed to run on small hardware, including microcontrollers, low-
   performance CPUs and AI chipsets.  This technology development
   direction is called On-device AI or TinyML[tinyML].

   In this document, we consider how to configure the network and system
   in terms of AI inference service to provide AI service in the IoT
   environment.  In the IoT environment, the technology of collecting
   sensing data from various sensors and delivering it to the cloud has
   already been studied by many standardization organizations including
   the IETF and many standards have been developed.  Now, after creating
   an AI model to provide AI services based on the collected data, how
   to configure this AI model as a system has become the main research
   goal.  Until now, it has been common to develop AI services that

Hong, et al.              Expires 25 April 2024                 [Page 3]
Internet-Draft            Deploying AI services             October 2023

   collect data and perform inferences from the trained servers, but in
   terms of the spread of AI services, it is not appropriate to use
   expensive servers to provide AI services.  In addition, since the
   server that collects and trains data mainly exists in the form of a
   cloud server, there are also many problems in proceeding in the form
   of requesting AI service by connecting a large number of terminals to
   these cloud servers to provide AI services.  Therefore, when an AI
   service is requested to an edge device located at a close distance,
   it may have effects such as real-time service support, network
   traffic reduction, and important data security rather than requesting
   an AI service to an AI server located in a distant
   cloud[I-D.irtf-t2trg-iot-edge].

   Even if an edge device is used to serve AI services, it is still
   important to connect to an AI server in the cloud for tasks that take
   a lot of time or require a lot of data.  Therefore, an offloading
   technique for properly distributing the workload between the cloud
   server and the edge device is also a field that is being actively
   studied.  In this contribution, in the following proposed network
   structure, the points to be considered in the environment where a
   client connects to a server and an edge device and requests an AI
   service are derived and described.  That is, the following
   considerations and options could be derived.

   *  AI inference service execution entity

   *  Hardware specifications of the machine to perform AI inference
      services

   *  Selection of AI models to perform AI inference services

   *  A method of providing AI services from cloud servers or edge
      devices

   *  Communication method to transmit data to request AI inference
      service

   The proposed considerations and items could be used to describe the
   use case of self-driving car and digital twin network.  Since
   providing AI services in a distributed method can provide various
   advantages, it is desirable to apply it to self-driving car and
   digital twin network.

Hong, et al.              Expires 25 April 2024                 [Page 4]
Internet-Draft            Deploying AI services             October 2023

2.  Procedure to provide AI services

   Since research on AI services has been started for a long time, there
   may be shapes to provide various types of AI services.  However, due
   to the nature of AI technology, in general, a system for providing AI
   services consists of the following steps [AI_inference_archtecture]
   [Google_cloud_iot].

+-----------+  +-----------+  +-----------+  +-----------+  +-----------+
| Collect & |  | Analysis &|  |   Train   |  |  Deploy & |  | Monitor & |
|  Store    |->| Preprocess|->|  AI model |->| Inference |->|  Maintain |
|   data    |  |    data   |  |           |  |  AI model |  |  Accuracy |
+-----------+  +-----------+  +-----------+  +-----------+  +-----------+
|<--------->|  |<------------------------>|  |<--------->|  |<--------->|
  Sensor, DB              AI Server              Target       AI Server &
                                                 machine    Target machine
|<---------------->|<--------------------->|<-------------->|<--------->|
      Interent              Local                Internet      Local &
                                                              Internet

                    Figure 1: AI service workflow

   *  Data collection & Store

   *  Data Analysis & Preprocess

   *  AI Model Training

   *  AI Model Deploy & Inference

   *  Monitor & Maintain Accuracy

   In the data collection step, data required for training is prepared
   by collecting data from sensors and IoT devices or by using data
   stored in a database.  Equipment involved in this step includes
   sensors, IoT devices and servers that store them, and database
   servers.  Since the operations performed at this step are conducted
   through the Internet, many IoT technologies studied by the IETF so
   far have developed technologies suitable for this step.

   In the data analysis and pre-processing step, the features of the
   prepared data are analyzed and pre-processing for training is
   performed.  Equipment involved in this step includes a high-
   performance server equipped with a GPU and a database server, and is
   mainly performed in a local network.

Hong, et al.              Expires 25 April 2024                 [Page 5]
Internet-Draft            Deploying AI services             October 2023

   In the model training step, a training model is created by applying
   an algorithm suitable for the characteristics of the data and the
   problem to be solved.  Equipment involved in this step includes a
   high-performance server equipped with a GPU, and is mainly performed
   on a local network.

   In the model deploying and inference service provision step, the
   problem to be solved (e.g., classification, regression problem) is
   solved using AI technology.  Equipment involved in this step may
   include a target machine, a client, a cloud, etc. that provide AI
   services, and since various equipment is involved in this stage, it
   is conducted through the Internet.  This document summarizes the
   factors to be considered at this step.

   In the accuracy monitoring step, if the performance deteriorates due
   to new data, a new model is created through re-training, and the AI
   service quality is maintained by using the newly created model.  This
   step is the same as described in the model training, model deploying,
   and inference service provision steps described in the previous step
   because re-training and model deploying are performed again.

3.  Network configuration structure to provide AI services

   In general, after training a AI model, the AI model can be built on a
   local machine for AI model deploying and inference services to
   provide AI services.  Alternatively, we can place AI models on cloud
   servers or edge devices and make AI service requests remotely.  In
   addition, for overall service performance, some AI service requests
   to the cloud server and some AI service requests to edge devices can
   be performed through appropriate load balancing.

3.1.  AI inference service on Local machine

   The following figure shows a case where a client module requesting AI
   service on the same local machine requests AI service from an AI
   server module on the same machine.

 +---------------------------------------------------------------------+
 |                                                                     |
 |   +-----------------+        Request AI      +-----------------+    |
 |   |  Client module  |    Inference service   |  Server module  |    |
 |   | for AI service  |----------------------->| for AI service  |    |
 |   |                 |<-----------------------|                 |    |
 |   +-----------------+        Reply AI        +-----------------+    |
 |                           Inference result                          |
 +---------------------------------------------------------------------+
                                  Local machine

Hong, et al.              Expires 25 April 2024                 [Page 6]
Internet-Draft            Deploying AI services             October 2023

            Figure 2: AI inference service on Local machine

   This method is often used when configuring a system focused on
   training AI models to improve the inference accuracy and performance
   of AI models without considering AI services or AI model deploying
   and inference in particular.  In this case, since the client module
   that requests the AI inference service and the AI server module that
   directly performs the AI inference service are on the same machine,
   it is not necessary to consider the communication/network environment
   or service provision method too much.  Alternatively, this method can
   be used when we want to simply decorate the AI inference service on
   one machine without changing the AI service in the future, such as an
   embedded machine or a customized machine.

   In this case, a high level of hardware performance is not required to
   train the AI model, but hardware performance sufficient to run the AI
   inference service is required, so it is possible on a machine with a
   certain amount of hardware performance.

3.2.  AI inference service on Cloud server

   The following figure shows the case where the client module that
   requests AI service and the AI server module that directly performs
   AI service run on different machines.

                                  +--------------------------------------+
+------------------------+        |     +---------------------------+    |
|   +-----------------+  |        |     |     +-----------------+   |    |
|   |   Client module |<-+--------+-----+---->|   Server module |   |    |
|   |  for AI service |  |        |     |     |  for AI service |   |    |
|   +-----------------+  |        |     |     +-----------------+   |    |
+------------------------+        |     + --------------------------+    |
       Local machine              |             Server machine           |
                                  +--------------------------------------+
                                                Cloud(Internet)

Hong, et al.              Expires 25 April 2024                 [Page 7]
Internet-Draft            Deploying AI services             October 2023

            Figure 3: AI inference service on Cloud server

   In this case, the client module requesting the AI inference service
   runs on the local machine, and the AI server module that directly
   performs the AI inference service runs on a separate server machine,
   and this server machine is in the cloud network.  In this case, the
   performance of the local machine does not need to be high because the
   local machine simply needs to request the AI inference service and,
   if necessary, deliver only the data required for the AI service
   request.  For the AI server module that directly performs AI
   inference service, we can set up our own AI server, or we can use
   commercial clouds such as Amazon, Microsoft, and Google.

3.3.  AI inference service on Edge device

   The following figure shows the case where the client module that
   requests AI service and the AI server module that directly performs
   AI service are separated, and the AI server module is located in the
   edge device.

                                  +--------------------------------------+
+------------------------+        |     +---------------------------+    |
|   +-----------------+  |        |     |     +-----------------+   |    |
|   |   Client module |<-+--------+-----+---->|   Server module |   |    |
|   |  for AI service |  |        |     |     |  for AI service |   |    |
|   +-----------------+  |        |     |     +-----------------+   |    |
+------------------------+        |     + --------------------------+    |
       Local machine              |                Edge device           |
                                  +--------------------------------------+
                                                  Edge network

            Figure 4: AI inference service on Edge device

   Even in this case, the client module that requests the AI inference
   service runs on the local machine, the AI server module that directly
   performs the AI inference service runs on the edge device, and the
   edge device is in the edge network.  Even in this case, the client
   module that requests the AI inference service runs on the local
   machine, the AI server module that directly performs the AI inference
   service runs on the edge device, and the edge device is in the edge
   network.  The AI module that directly performs the AI inference
   service on the edge device can directly configure the edge device or
   use a commercial edge computing module.

Hong, et al.              Expires 25 April 2024                 [Page 8]
Internet-Draft            Deploying AI services             October 2023

   The difference from the above case where the AI server module is in
   the cloud is that the edge device is usually close to the client,
   whereas the performance is lower than that of the server in the
   cloud, so there are advantages in data transfer time and inference
   time, but in unit time Inference service performance is poor.

3.4.  AI inference service on Cloud server and Edge device

   The following figure shows the case where AI server modules that
   directly perform AI services are distributed in the cloud and edge
   devices.

                                  +--------------------------------------+
+------------------------+        |     +---------------------------+    |
|   +-----------------+  |        |     |     +-----------------+   |    |
|   |   Client module |<-+---+----+-----+---->|   Server module |   |    |
|   |  for AI service |<-+---+    |     |     |  for AI service |   |    |
|   +-----------------+  |   |    |     |     +-----------------+   |    |
+------------------------+   |    |     + --------------------------+    |
       Local machine         |    |                Edge device           |
                             |    +--------------------------------------+
                             |                    Edge network
                             |
                             |    +--------------------------------------+
                             |    |     +---------------------------+    |
                             |    |     |     +-----------------+   |    |
                             +----+-----+---->|   Server module |   |    |
                                  |     |     |  for AI service |   |    |
                                  |     |     +-----------------+   |    |
                                  |     + --------------------------+    |
                                  |              Server machine          |
                                  +--------------------------------------+
                                                 Cloud(Internet)

    Figure 5: AI inference service on Cloud sever and Edge device

   There is a difference between the AI server module performed in the
   cloud and the AI server module performed on the edge device in terms
   of AI inference service performance.  Therefore, the client
   requesting the AI inference service may request by distributing the
   AI inference service request to the cloud and edge device
   appropriately in order to perform the desired AI service.  In other
   words, in the case of an AI service with low inference accuracy but
   short inference time, we can request an AI inference service to the
   edge device.

Hong, et al.              Expires 25 April 2024                 [Page 9]
Internet-Draft            Deploying AI services             October 2023

3.5.  AI inference service on horizontal multiple servers

   In the previous section, to provide AI inference service, the network
   configuration that consisted of local machines, edge devices, and
   cloud servers is a kind of vertical hierarchy.  Because the
   capabilities of each machine are different, the overall performance
   of the network using vertical hierarchy is dependent of each machine.
   Generally, a cloud server has a most powerful performance and then an
   edge device has the second powerful performance.

   In this network configuration, AI service may have different
   performance according to the load level of the server, computing
   capability of the server machine and link-state between the local
   machine and the server machines of the horizontal level.  Thus, to
   look for the server machine that can support the best AI service, it
   is necessary for the network element that can monitor network link-
   state and current state of the computing capability of the server
   machines and the network load-balance that can perform a scheduling
   policy of load balancing.  The following figure shows the case where
   the local machine that requests AI service to horizontal multiple
   cloud servers.

Hong, et al.              Expires 25 April 2024                [Page 10]
Internet-Draft            Deploying AI services             October 2023

                                  +--------------------------------------+
                                  |     +---------------------------+    |
                                  |     |     +-----------------+   |    |
                             +----+-----+---->|   Server module |   |    |
                             |    |     |     |  for AI service |   |    |
                             |    |     |     +-----------------+   |    |
                             |    |     + --------------------------+    |
                             |    |              Server machine 1        |
                             |    +--------------------------------------+
                             |                   Cloud(Internet)
                             |
                             |    +--------------------------------------+
+------------------------+   |    |     +---------------------------+    |
|   +-----------------+  |   |    |     |     +-----------------+   |    |
|   |   Client module |<-+---+----+-----+---->|   Server module |   |    |
|   |  for AI service |<-+---+    |     |     |  for AI service |   |    |
|   +-----------------+  |   |    |     |     +-----------------+   |    |
+------------------------+   |    |     + --------------------------+    |
       Local machine         |    |              Server machine 2        |
                             |    +--------------------------------------+
                             |                   Cloud(Internet)
                             |
                             |    +--------------------------------------+
                             |    |     +---------------------------+    |
                             |    |     |     +-----------------+   |    |
                             +----+-----+---->|   Server module |   |    |
                                  |     |     |  for AI service |   |    |
                                  |     |     +-----------------+   |    |
                                  |     + --------------------------+    |
                                  |              Server machine 3        |
                                  +--------------------------------------+
                                                 Cloud(Internet)

    Figure 6: AI inference service on horizontal multiple servers

3.6.  Network-side utilization for AI learning

   Collecting and preprocessing of data and training an AI model
   requires a high-performance resource such as CPU, GPU, Power, and
   Storage.  To mitigate this requirement, we can utilize a network-side
   configuration.  Typically, federating learning is a machine learning
   technique that trains an AI model across multiple decentralized
   servers.  It is a contrast to traditional centralized machine
   learning techniques where all the local datasets are uploaded to one
   server.  In this federated learning, it enables multiple network
   nodes to build a common machine learning model.

Hong, et al.              Expires 25 April 2024                [Page 11]
Internet-Draft            Deploying AI services             October 2023

   And, transfer learning is a machine learning technique that focuses
   on storing information gained while solving one problem and applying
   it to a different but related problem.  In this transfer learning, we
   can utilize a network configuration to transfer common information
   and knowledge between different network nodes.

4.  Considerations for configuring a network to provide AI services

   As described in the previous chapter, the AI server module that
   directly performs AI inference services by utilizing AI models can be
   performed on a local machine or a cloud server or an edge device.

   In theory, if AI inference service is performed on a local machine,
   AI service can be provided without communication delay time or packet
   loss, but a certain amount of hardware performance is required to
   perform AI service inference.  So, in the future environment where AI
   services become popular, such as when various AI services are
   activated and AI services are disseminated, the cost of a machine
   that performs AI services is important

   If so, whether the AI inference service will be performed on the
   cloud server or the discount price on the edge device can be a
   determining factor in the system configuration.

4.1.  Considerations according to the functional characteristics of the
      hardware

   When AI inference service request is made to a distant cloud server,
   it may take a lot of time to transmit, but it has the advantage of
   being able to perform many AI inference service requests in a short
   time, and the accuracy of AI service inference increases.
   Conversely, when an AI service request is made to a nearby edge
   device, the transmission time is short, but many AI inference service
   requests cannot be performed at once, and the accuracy of AI service
   inference is lowered.

   Therefore, by analyzing the characteristics and requirements of the
   AI service to be performed, it is necessary to determine where to
   perform the AI inference service on a local machine, a cloud server,
   or an edge device.

   The hardware characteristics of the machine performing the AI service
   varies.  In general, machines on cloud servers are viewed as machines
   with higher performance than edge devices.  However, the performance
   of AI inference service varies depending on how the hardware such as
   CPU, RAM, GPU, and network interface is configured for each cloud
   server and edge device.  If we do not think about cost, it is good to
   configure a system for performing AI services with a machine with the

Hong, et al.              Expires 25 April 2024                [Page 12]
Internet-Draft            Deploying AI services             October 2023

   best hardware performance, but in reality, we should always consider
   the cost when configuring the system.  So, according to the
   characteristics and requirements of the AI service to be performed,
   the performance of the local machine, cloud server, and edge device
   must be determined.

   Performance evaluation is possible through the performance matrix
   presented in the standard of ETSI[MEC.IEG006].  The performance
   metrics suggested by the ETSI standard are as follows.  These metrics
   is divided into two groups, namely Functional metrics, which assess
   the user performance and include some classical indexes such as
   latency in task execution, device energy efficiency, bit-rate, loss
   rate, jitter, Quality of Service (QoS), etc.; and Non-functional
   metrics that, instead, focus on the MEC(Mobile Edge Computing)
   network deployment and management.  Non-functional metrics include
   the following indexes.  Service life-cycle(instantiation, service
   deployment, service provisioning, service update (e.g. service
   scalability and elasticity), service disposal), service availability
   and fault tolerance (aka reliability), service processing/
   computational load, global mobile equipment host load, number of API
   request (more generally number of events) processed/second on mobile
   equipment host, delay to process API request (north and south),
   number of failed API request.  The sum of service instantiation,
   service deployment, and service provisioning provide service boot-
   time.

4.2.  Considerations according to the characteristics of the AI model

   According to the characteristics of the AI service, although not
   directly related to communication/network, the biggest influence on
   AI inference services is the AI model to be used for AI inference
   service.  For example, in AI services such as image classification,
   there are various types of AI models such as ResNet, EfficientNet,
   VGG, and Inception.  These AI models differ in AI inference accuracy,
   but also in AI model file size and AI inference time.  AI models with
   the highest inference accuracy typically have very large file sizes
   and take a lot of AI inference time.  So, when constructing an AI
   service system, it is not always good to choose an AI model with the
   highest AI inference accuracy.  Again, it is important to select an
   AI model according to the characteristics and requirements of the AI
   service to be performed.

   Experimentally, it is recommended to use an AI model with high AI
   inference accuracy in the cloud server, and use an AI model that can
   provide fast AI inference service although the AI inference accuracy
   is slightly lower for the fast AI inference service in the edge
   device.

Hong, et al.              Expires 25 April 2024                [Page 13]
Internet-Draft            Deploying AI services             October 2023

   It might be a bit of an implementation issue, but we should also
   consider how we deliver AI services on cloud servers or edge devices.
   With the current technology, a traditional web server method or a
   server method specialized for AI service inference (e.g., Google's
   Tensorflow Serving) can be used.  Traditional web server methods such
   as Flask and Django have the advantage of running on various types of
   machines, but since they are designed to support general web
   services, the service execution time is not fast.  Tensorflow Serving
   uses the features of Tensorflow to make AI service inference services
   very fast and efficient.  However, older CPUs that do not support AVX
   cannot use the Tensorflow serving function because Google's
   Tensorflow does not run.  Therefore, rather than unconditionally
   using the server method specialized in AI service inference, it is
   necessary to decide the AI server module method that provides AI
   services in consideration of the hardware characteristics of the AI
   system that can be built.

4.3.  Considerations according to the characteristics of the
      communication method

   The communication method for transferring data to request AI
   inference service is also an important decision in constructing an AI
   system.  Using the traditional REST method, it can be used for
   various machines and services, but its performance is inferior to
   Google's gRPC.  There are many advantages to using gRPC for AI
   inference services because Google's gRPC enables large-capacity data
   transfer and efficient data transfer compared to REST.

   Cloud-edge collaboration-based AI service development is actively
   underway.  In particular, in the case of AI services that are
   sensitive to network delays, such as object recognition and
   autonomous vehicle services, (micro)services for inference are placed
   on edge devices to obtain fast inference results and provide
   services.  As such, in the development of intelligent IoT services,
   various devices that can provide computing services within the
   network, such as edge devices, are being added as network elements,
   and the number of IoT devices using them is rapidly increasing.
   Therefore, a new function for computing resource management and
   operation is required in terms of providing computing services within
   the network.

5.  Use cases of deploying AI services in a distributed method

Hong, et al.              Expires 25 April 2024                [Page 14]
Internet-Draft            Deploying AI services             October 2023

5.1.  Deploying AI services in Self-driving car

   Various sensors are used in self-driving cars, and the final judgment
   is made by combining these data.  Among them, camera data-based
   object detection solves parts that expensive equipment such as LiDAR
   and RADAR cannot solve.  Camera-based object detection performs
   various tasks, and in addition to lane recognition for maintaining
   driving lanes and changing lanes, it also supports safe driving and
   parking assistance by distinguishing shape information such as
   pedestrians, signs, and parking vehicles along the road.

   In order to perform such driving assistance and autonomous driving,
   object detection needs to be performed in real time.  The minimum
   FPS(Frames Per Second) to be considered real-time in autonomous
   driving is 30 FPS[Object_detection].  No matter how high the accuracy
   is, it cannot be used for autonomous driving if it does not meet the
   corresponding reference value.

   Task offloading refers to a technology or structure that transfers
   computing tasks to other processing devices or systems to perform
   them.  Task offloading can quickly process tasks that exceed the
   performance limits of devices that lack resources by delivering tasks
   from devices with limited computing power, storage space, and power
   to devices that are rich in computing resources.

   For devices with low hardware performance (e.g., NVIDIA Jetson Nano
   board, Qual-core ARM A57, 4GB RAM), all locally without task
   offloading results in 4.6 FPS, which is difficult to perform object
   detection-based autonomous driving.  On the other hand, if task
   offloading is applied to perform object detection on devices with
   high hardware performance (e.g., Intel i7, RTX 3060, 32GB RAM) and
   the rest of the work is performed on the client, 41.8 FPS will be
   obtained.  This is a result that satisfies 30 FPS, which is the
   reference FPS of object detection-based autonomous driving.

   In the case of AI services such as object detection, if it is
   difficult to perform on resource-constrained devices, it can be seen
   that the task offloading structure shows some efficiency.  However,
   without performing all operations locally, task offloading operations
   between network nodes can affect the entire time because the larger
   the size of the data, the greater the communication latency.
   Therefore, in such a network distributed environment, the provision
   of AI services should be designed in consideration of various
   variables.  The Figure 7 shows an example of distributed AI
   deployment in a self-driving car when a car does not have enough
   capabilities to proceed the object detection operation in real-time
   and it asks some tasks to edge devices and cloud servers.

Hong, et al.              Expires 25 April 2024                [Page 15]
Internet-Draft            Deploying AI services             October 2023

                                  +--------------------------------------+
+------------------------+        |     +---------------------------+    |
|   +-----------------+  |        |     |     +-----------------+   |    |
|   |Object detection |<-+---+----+-----+---->|Object detection |   |    |
|   |     service     |<-+---+    |     |     |     service     |   |    |
|   +-----------------+  |   |    |     |     +-----------------+   |    |
+------------------------+   |    |     + --------------------------+    |
           Car               |    |                Edge device           |
                             |    +--------------------------------------+
                             |                    Edge network
                             |
                             |    +--------------------------------------+
                             |    |     +---------------------------+    |
                             |    |     |     +-----------------+   |    |
                             +----+-----+---->|Object detection |   |    |
                                  |     |     |     service     |   |    |
                                  |     |     +-----------------+   |    |
                                  |     + --------------------------+    |
                                  |              Server machine          |
                                  +--------------------------------------+
                                                 Cloud(Internet)

  Figure 7: Distributed object detection service in self-driving car

5.2.  Deploying AI services in Digital twin network

   Digital twin networks also need to build distributed AI services.
   The purpose of a digital twin network is described in
   [I-D.irtf-nmrg-network-digital-twin-arch].  In particular, the
   digital twin network provides network operators with technology that
   enables stable operation of the physical network and stable execution
   of optimal network policies and deployment procedures.  To achieve
   this, the digital twin network will use AI capabilities for various
   purposes.

   Various AI functions will be applied for optimal network operation
   and management.  However, the actual physical network consists of
   many network devices and has a complex structure.  In addition, in a
   large-scale network environment, the network overhead is very large
   to collect and store information from many network devices in a
   centralized manner, and to create and operate network operation
   policies based on it.

   Therefore, there is a need for a method to apply AI functions based
   on a distributed form for network operation and management.  In
   particular, the actual physical network structure is built in a

Hong, et al.              Expires 25 April 2024                [Page 16]
Internet-Draft            Deploying AI services             October 2023

   logical hierarchical structure.  Therefore, it is necessary to apply
   a distributed AI method that considers the logical hierarchical
   network structure environment.

   In order to optimally perform network operation and management
   through distributed AI methods, it is necessary to generate AI
   function-based network operation and management policy models and an
   operational method to distribute the generated AI function-based
   network policies.  In particular, in order to operate a digital twin
   network in a large-scale network environment, it is necessary to
   generate AI-based network policy models in a distributed manner.  A
   federated learning algorithm or a transfer learning algorithm that
   can learn large-scale networks in a distributed manner can be
   applied.

          +-----------------------------------------------------+
          |                                                     |
          |        Distributed netwrok learning model           |
          |        in large-scale network environment           |
          |                                                     |
          |            +-------------+------------+             |
          |            |          Master          |             |
          |            | (AI based Policy model)  |             |
          |            +-------------+------------+             |
          |                          |                          |
          |        +-----------+-----+-----+-----------+        |
          |        |           |           |           |        |
          |   +----+----+ +----+----+ +----+----+ +----+----+   |
          |   |  Worker | |  Worker | |  Worker | |  Worker |   |
          |   | (Agent) | | (Agent) | | (Agent) | | (Agent) |   |
          |   +----+----+ +----+----+ +----+----+ +----+----+   |
          |        |           |           |           |        |
          |   +----+----+ +----+----+ +----+----+ +----+----+   |
          |   |  Local  | |  Local  | |  Local  | |  Local  |   |
          |   |  Data   | |  Data   | |  Data   | |  Data   |   |
          |   |  Repo-  | |  Repo-  | |  Repo-  | |  Repo-  |   |
          |   |  sitory | |  sitory | |  sitory | |  sitory |   |
          |   +---------+ +---------+ +---------+ +---------+   |
          |                                                     |
          +-----------------------------------------------------+

Hong, et al.              Expires 25 April 2024                [Page 17]
Internet-Draft            Deploying AI services             October 2023

        Figure 8: Distributed learning model of network learning for
                            Digital twin network

   As shown in Figure 8, in order to learn a large-scale network through
   a distributed learning method, a local data repository to store
   network data must be established in each region, for example, based
   on location or AS (Autonomous System).  Therefore, the distributed
   learning method learns through each worker (agent) based on the local
   network data stored in the local network data repository, and
   generates a large-scale network policy model through the master.
   This distributed learning method can reduce the network overhead of
   centralized data collection and storage, and reduce the time required
   to create AI models for network operation and management policies for
   large-scale networks.  In addition, the network policy model
   generated by the worker can be used as a locally optimized network
   policy model to provide AI-based network operation and management
   policy services optimized for local network operations.

   The distributed deployment of trained AI network policy models can be
   deployed on network devices that can manage and operate the local
   network to minimize network data movement.  For example, in a large-
   scale network consisting of multiple ASes, AI network policy models
   can be deployed per AS to optimize network operation and management.
   Figure 9 shows an example of operating and managing a network by
   distributing AI network policy models by AS.

        +---------------------------------------------------------+
        |              +-------------+------------+               |
        |              |          Master          |               |
        |              | (AI-based Network Policy |               |
        |              |     model management)    |               |
        |              +-------------+------------+               |
        |                            |                            |
        |              +-------------+-------------+              |
        |     Worker   |                 Worker    |              |
        |   +----------+----------+     +----------+----------+   |
        |   |    Network device   |     |    Network device   |   |
        |   |  (AI-based network  |     |  (AI-based network  |   |
        |   |     Policy model)   |     |     Policy model)   |   |
        |   +----------+----------+     +----------+----------+   |
        |              |                           |              |
        |   +----------+----------+     +----------+----------+   |
        |   |        AS_1         |     |        AS_2         |   |
        |   +---------------------+     +---------------------+   |
        +---------------------------------------------------------+

Hong, et al.              Expires 25 April 2024                [Page 18]
Internet-Draft            Deploying AI services             October 2023

    Figure 9: Distributed deployment of trained AI network policy models

6.  IANA Considerations

   There are no IANA considerations related to this document.

7.  Security Considerations

   When AI service is performed on a local machine, there is no security
   issue, but when AI service is provided through a cloud server or edge
   device, IP address and port number may be known to the outside can
   attack.  Therefore, when providing AI services by utilizing machines
   on the network such as cloud servers and edge devices, it is
   necessary to analyze the characteristics of the modules to be used
   well, identify vulnerabilities in security, and take countermeasures.

8.  Acknowledgements

   TBA

9.  Informative References

   [RFC6574]  Tschofenig, H. and J. Arkko, "Report from the Smart Object
              Workshop", RFC 6574, DOI 10.17487/RFC6574, April 2012,
              <https://www.rfc-editor.org/info/rfc6574>.

   [RFC7452]  Tschofenig, H., Arkko, J., Thaler, D., and D. McPherson,
              "Architectural Considerations in Smart Object Networking",
              RFC 7452, DOI 10.17487/RFC7452, March 2015,
              <https://www.rfc-editor.org/info/rfc7452>.

   [I-D.irtf-t2trg-iot-edge]
              Hong, J., Hong, Y., de Foy, X., Kovatsch, M., Schooler,
              E., and D. Kutscher, "IoT Edge Challenges and Functions",
              Work in Progress, Internet-Draft, draft-irtf-t2trg-iot-
              edge-10, 15 September 2023,
              <https://datatracker.ietf.org/doc/html/draft-irtf-t2trg-
              iot-edge-10>.

   [I-D.irtf-nmrg-network-digital-twin-arch]
              Zhou, C., Yang, H., Duan, X., Lopez, D., Pastor, A., Wu,
              Q., Boucadair, M., and C. Jacquenet, "Digital Twin
              Network: Concepts and Reference Architecture", Work in
              Progress, Internet-Draft, draft-irtf-nmrg-network-digital-
              twin-arch-03, 27 April 2023,
              <https://datatracker.ietf.org/doc/html/draft-irtf-nmrg-
              network-digital-twin-arch-03>.

Hong, et al.              Expires 25 April 2024                [Page 19]
Internet-Draft            Deploying AI services             October 2023

   [CG-AIoT]  "ITU-T CG-AIoT", <https://www.itu.int/en/ITU-T/
              studygroups/2017-2020/20/Pages/ifa-structure.aspx>.

   [tinyML]   "tinyML Foundation", <https://www.tinyml.org/>.

   [AI_inference_archtecture]
              "IBM Systems, AI Infrastructure Reference Architecture",
              <https://www.ibm.com/downloads/cas/W1JQBNJV>.

   [Google_cloud_iot]
              "Bringing intelligence to the edge with Cloud IoT",
              <https://cloud.google.com/blog/products/gcp/bringing-
              intelligence-edge-cloud-iot>.

   [MEC.IEG006]
              ETSI, "Mobile Edge Computing; Market Acceleration; MEC
              Metrics Best Practice and Guidelines", Group
              Specification ETSI GS MEC-IEG 006 V1.1.1 (2017-01),
              January 2017.

   [Object_detection]
              Lewis, "Object Detection for Autonomous Vehicles Gene",
              2016.

Authors' Addresses

   Yong-Geun Hong
   Daejeon University
   62 Daehak-ro, Dong-gu
   Daejeon
   34520
   South Korea
   Phone: +82 42 280 4841
   Email: yonggeun.hong@gmail.com

   SeokBeom Oh
   KSA
   Digital Transformation Center, 5
   Teheran-ro 69-gil, Gangnamgu
   Seoul
   06160
   South Korea
   Phone: +82 2 1670 6009
   Email: isb6655@korea.ac.kr

Hong, et al.              Expires 25 April 2024                [Page 20]
Internet-Draft            Deploying AI services             October 2023

   Joo-Sang Youn
   DONG-EUI University
   176 Eomgwangno Busan_jin_gu
   Busan
   614-714
   South Korea
   Phone: +82 51 890 1993
   Email: joosang.youn@gmail.com

   SooJeong Lee
   Korea University/KT
   2511 Sejong-ro
   Sejong City
   30019
   South Korea
   Email: ngenius@korea.ac.kr

   Seung-Woo Hong
   ETRI
   218 Gajeong-ro Yuseong-gu
   Daejeon
   34129
   South Korea
   Phone: +82 42 860 1041
   Email: swhong@etri.re.kr

   Ho-Sun Yoon
   ETRI
   218 Gajeong-ro Yuseong-gu
   Daejeon
   34129
   South Korea
   Phone: +82 42 860 5329
   Email: yhs@etri.re.kr

Hong, et al.              Expires 25 April 2024                [Page 21]