Internet DRAFT - draft-zhang-rtgwg-srv6-computing-connect-usecases
draft-zhang-rtgwg-srv6-computing-connect-usecases
Network Working Group X. Zhang
Internet-Draft F. Yang
Intended status: Standards Track W. Cheng
Expires: April 20, 2023 China Mobile
Z. Fu
New H3C Technologies
October 20, 2023
Usecases of SRv6 Based Computing Interconnection Network
draft-zhang-rtgwg-srv6-computing-connect-usecases-03
Abstract
The requirements of computing interconnection are increasingly
attracting the attention of service providers. They have been thinking
about how to leverage their network advantages to provide integrated
networking and computing services. This document describes some
scenarios of using SRv6 based network technology which can partially
meet the service requirement of computing interconnection.
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 20 April 2023.
Copyright Notice
Copyright (c) 2022 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.
Zhang, et al. Expires 20 April 2023 [Page 1]
Internet-Draft SRv6 Based Computing interconnection Usecases April 2022
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2. Usage Scenarios of SRv6 based computing interconnection network . 3
2.1. SRv6 based computing interconnection network architecture . . 3
2.2. Path scheduling . . . . . . . . . . . . . . . . . . . . . . 5
2.3. Resource isolation . . . . . . . . . . . . . . . . . . . . . 5
2.4. Multi segment path orchestration . . . . . . . . . . . . . . 5
2.5. Multi Service orchestration . . . . . . . . . . . . . . . . 6
2.6. Network reliability for computing interconnection . . . . . . 6
2.7. Application-aware networking . . . . . . . . . . . . . . . . 6
2.8. Operations, Administration and Maintenance . . . . . . . . . 7
3. Collaboration between computing and network . . . . . . . . . . . 7
4. Best practices. . . . . . . . . . . . . . . . . . . . . . . . . . 8
5. Security Considerations . . . . . . . . . . . . . . . . . . . . . 9
6. IANA Considerations . . . . . . . . . . . . . . . . . . . . . . . 9
7. Normative References . . . . . . . . . . . . . . . . . . . . . . 9
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . . 9
1. Introduction
With the advent of new technology such as cloud computing, big data,
artificial intelligence, etc., the demand for computing resource is
continuously increasing. More and more data centers, intelligent
computing centers, and supercomputing centers have been built to meet
the growing demand of computing resource. Usually, these computing
centers are centralized(e.g. central cloud). Especially, in some
emerging industries, such as self-driving, cloud AR/VR, telemedicine,
etc., there are not only requirements for computing resource, but also
for quick delivery and guaranteed quality. These requirements are
usually related to network factors, such as delay, bandwidth, and
jitter, etc. These services can be deployed not only on the central
cloud, but also on distributed edge nodes.
In order to coordinate computing resource at different levels(center,
edge and end) uniformly, and to meet user's requirements for computing
power and network, a new type of network is proposed which can combine
computing and network information and provide optimal allocation,
association and scheduling of computing, storage and network resources.
We call it computing interconnection network in this document. The
computing interconnection network is a converged architecture with
computing and network.
The computing interconnection network has attracted the attention of
many service providers. Lots of service providers have proposed their
own concepts of computing interconnection network, and have also
released relevant technical white papers. Different computing resources
are interconnected through the network. The goal of computing
interconnection network is to achieve "ubiquitous computing resource,
computing network symbiosis, intelligent orchestration, and integrated
Zhang, et al. Expires 20 April 2023 [Page 2]
Internet-Draft SRv6 Based Computing interconnection Usecases April 2022
services", and gradually develop into a infrastructure-level service
that can be "once connected, use anywhere" like water and electricity.
These goals present some challenges to current network architecture.
Segment Routing Architecture [RFC8402] proposes a network paradigm
based on a source routing mechanism. The segment routing network has
the remarkable characteristics of simplifying the control plane and
network state. In addition, segment routing can program the network
functions that need to be performed along the way, so that the packets
can be transmitted and processed in the desired way. Segment routing
can be applied to IPv6 network, which is called SRv6. In addition to
inheriting the advantages of source routing, SRv6 has many other
advantages. Firstly, IPv6 can provide more addresses to meet the needs
of the Internet of Things. Secondly, SRv6 has three levels of
programmability that are extremely scalable. Finally, SRv6 also
supports in-suit OAM(IOAM), service chain, slicing and other features.
SRv6 is the trend in the evolution of IP networks to intelligent IP
networks.
The network for computing interconnection has attracted the attention
of many service providers. They also have deployed new bearer network
with SRv6 to provide better connection service. Based on the flexible
scalability, programmability, simplicity and other advantages, SRv6
can meet some requirements of computing interconnection network. This
document introduces some usage scenarios of SRv6 based computing
interconnetion network.
2. Usage Scenarios of SRv6 based computing interconnection network
2.1 SRv6 based computing interconnection network architecture
The following figure shows a typical architecture of SRv6 based
computing interconnection network. There are two layers here, one is
the infrastructure layer and the other is the control layer.
*Infrastructure layer:
Edge: The network edge device of computing interconnection. In this
document, Edge is both the edge device for computing
interconnection and the endpoint of the SRv6 path.
Computing Resource: The computing resources connected to computing
interconnection Edge. It can be cloud, edge or terminal and so on.
Client: Clients requesting computing services.
*Control layer:
Computing and Network Controller: Computing resource scheduling,
orchestration and network policy distribution, in this document,
CNC is used to represent it.
Zhang, et al. Expires 20 April 2023 [Page 3]
Internet-Draft SRv6 Based Computing interconnection Usecases April 2022
/-----------------------------------/
/ /
/ Computing and Network Controller /
/ (CNC) /
/-----------:--------:--------:-----/
: : :
: : :
...............................: : :.............
+----:-----+ : :
+----------+ | : :
|Computing | | : :
|Resource 1|-+ /-----------------------------:-----------------/ :
+----+-----+ / +--------------+ / :
| / +---------+ | SRv6 | +---------+ / :
+--------+--| Edge1 |--|Infrastructure|--| Edge3 |---+---+ :
/ +---------+ | | +---------+ / | :
/ | +--------------+ | / | :
/ | | | / | :
/ | +----------+ | / +-----+--:-+
/ | | Edge 2 | | /+----------+ |
/ | +----------+ | / |Computing | |
/ | | | / |Resource 2|-+
/--------------+---------------+--------------+-/ +----------+
| | |
+---+--+ +---+--+ +---+--+
+------+ | +------+ | +------+ |
|client|-+ |client|-+ |client|-+
+------+ +------+ +------+
Figure 1: SRv6 based computing interconnection network architecture
2.2 Path scheduling
when a computing request comes, it is necessary to decide which remote
computing node is available to provide the service. Both computing
power and network need to be considered for the decision. After the
computing node is determined, a SRv6 path fulfilling the SLA
requirement can be established to steers the packet to the destination,
i.e. the remote computing node. SRv6 is based on source routing
mechanism, which can compose the path information at the ingress of the
network. The path information is encapsulated in the packet, and
identified by a list of SIDs. Then, the packet only need to process the
outermost SID downstream. The downstream nodes on the forwarding path
can be stateless. SRv6 paths could be established according to default
metrics (e.g. cost) or user's policy. These paths can be in loose or
strict manner.
Zhang, et al. Expires 20 April 2023 [Page 4]
Internet-Draft SRv6 Based Computing interconnection Usecases April 2022
2.3 Resource isolation
The different services for computing interconnection will
be implemented on the same physical network. Some are sensitive to
network delay, such as AR/VR. Some are sensitive to packet loss, such
as storage services. Therefore, different services share the same
physical network, but they need to be isolated from each other.
This requires the network with slicing capabilities. Each slice is a
logical network. Different slices can provide the services with
different SLA requirements which are isolated from each other. SRv6's
programmability and protocol simplification could provide slicing
capabilities. Using the programmability of SRv6, network devices can
assign a specific SID and reserve hardware resources for each slice.
The device identifies the network slice based on the specific SID, and
steers the packet according to the topology and resources defined by
the slice. Then the packets with different SLA requirements can be
forwarded in different slices to meet the requirements of business
isolation.
2.4 Multi segment path orchestration
As described in chapter 1, the computing interconnection network is a
converged architecture with computing and network. The computing
resources is interconnected through the wide area network. The network
may be hierarchical, for example, including access, metro, backbone.
For each computing request, CNC needs to learn the state of the network
and computing resource comprehensively and make decisions according to
the user's constraint. The final selected computing node may cross
multiple autonomous domains. Users may want to obtain a link with low
delay, high bandwidth, or high reliability. Therefore, it is necessary
to consider how to obtain a path that meets SLA when spanning multiple
autonomous domains.
The SRv6's BSID realizes the opening of network capabilities.
Specifically, the SRv6 network identifies some paths which have
specific SLA metrics with a SID, such as low latency. The SID is called
BSID, which could be opened to CNC. CNC can select the appropriate BSID
according to the user's requirements for network. The BSID hides the
complex path information, and only one SID is presented externally.
The path represented by the BSID can be a complete path or a certain
segment of a complete path. Using SRv6's BSID, underlay and overlay can
be combined, and multiple domains can also be connected. SRv6 based
computing interconnection network can orchestrate network paths more
conveniently and concisely.
2.5 Multi service orchestration
In some scenarios, the computing service may need to pass through
multiple computing service nodes. For example, in order to meet the
requirements of user service's security and stability, when data
packets are transmitted in the network, they often need to pass through
various service nodes in sequence, such as firewalls, IPS, and
Zhang, et al. Expires 20 April 2023 [Page 5]
Internet-Draft SRv6 Based Computing interconnection Usecases April 2022
application accelerators. This can be achieved through the SRv6 service
function chain(SFC for short). SRv6 SFC is realized through the
programmability of SRv6. SFC uses specific SIDs to represent the
value-added services. The CNC can encode the value-added service
functions from the service request in the network path segment list by
SIDs, and forward and process the value-added service functions along
the path. This maximizes the ability to integrate the computing and
network services.
2.6 Network reliability for computing interconnection
Different applications hosted on the computing interconnect network
have different requirements for network reliability. Network failure
usually represents packets loss. However, many interactive multimedia
applications (such as cloud gaming) are very sensitive to packet loss,
dozens of milliseconds of packet loss will lead to a rapid decline in
the quality of service. The traditional fast rerouting mechanism of
IP network has some problems, such as complex configuration, worse
handover performance, etc. However, the computing interconnect network
based on SRv6 can solve these problems. For example, the Topology
Independent Loop-Free Alternate(TI-LFA) technology can provide
end-to-end local protection mechanism, which complete path switching
in a very short period of time after network failure. In addition,
SRv6 also provides a micro ring prevention mechanism, which prevent
traffic loop in a short time after fault recovery. Therefore, the
computing capacity can be continuously and stably transmitted.
2.7 Application-aware networking
Traditionally, applications and networks are separated, and the network
can only identify applications by means of five tuples. This method has
a coarser granularity and cannot understand the real needs of
applications for computing resource and networks. In computing
interconnection network, in order to provide efficient and quality
guaranteed computing services, the edge of the network is required to
identify different applications and their needs through incoming
packets, so as to provide different SLA services. Application-aware
networking for IPv6/SRv6 can meet this demand which can carry the
application identification and requirements for network and computing,
for example by using IPv6 extend head. Then the network edge can
perceive these applications and corresponding requirements, so as to
steer the packet to the appropriate SRv6 path.
2.8 Operations, Administration and Maintenance
As an infrastructure that can serve various industry customers or
individual users, the operation, administration and maintenance of
computing interconnection network is very important. The network
usually changes more frequently. When network quality deteriorates,
computing interconnection network needs to respond quickly and provide
a better path. Therefore, real-time monitoring of the current network
state is required, which can be used as the basis for more accurate
Zhang, et al. Expires 20 April 2023 [Page 6]
Internet-Draft SRv6 Based Computing interconnection Usecases April 2022
and reasonable scheduling decisions and guarantee the SLA requirements.
SRv6 supports in-situ OAM(IOAM), which can detect the network quality
in real-time and accurate way. Based on the real-time network status,
SRv6 based network can better serve computing scheduling
and network SLA guarantee.
3. Collaboration between computing and network
The core concept of computing interconnection network is collaboration
between computing and network. Computing and network are no longer
isolated entities, and they need to cooperate with each other.
Computing resources and network resources need to be managed and
allocated from a global perspective. The reference architecture given
in figure 1 is a possible implementation. In this architecture the
centralized CNC can realize the integrated orchestration, control and
management of the computing and network. The integrated orchestration
of computing and network which is aimed at the diversified and
customized requirements of computing and network convergence, could
design product and service models based on the flexible combination of
the atomic capability of the computing resource and network, and realize
the unified orchestration, deployment and guarantee of computing and
network services. The collaborative orchestration and scheduling of
computing and network provides a new network paradigm to accelerate
the digital transformation of society.
4. Best practices
Based on the above-mentioned important role of SRv6 in computing
interconnection network, as a typical practice, China Mobile has built
the infrastructure of SRv6 based DCI network and smart SD-WAN. The
SRv6-based DCI network can uniformly access various computing resources
and provide computing services. Smart SD-WAN is a new generation of
SD-WAN that integrates overlay and underlay networks. SRv6 based DCI
network and smart SD-WAN enhance coordination ability between computing
and network resource. It enable full connectivity of end, edge, cloud
and network, and combine user's intent to achieve collaborative
scheduling among application, computing and network, and improve
service quality assurance capabilities, to realize end-to-end,
differentiated, deterministic, and value-added computing network
services.
The following is a specific application example. Considering a CDN
scheduling system, CDN applications can be regarded as computing
services required by users. In the traditional scheduling mode,
scheduling system usually allocate CDN node according to the user's
geographic location. This will lead to a large number of users in a
hotspot area being assigned to the same CDN node, so that some CDN
nodes are busy, while others are very idle. This results in low
resource utilization, and service quality cannot be guaranteed. In
the computing interconnection network, CNC can manage computing resouce
and network resources at the same time. It can assign users in the same
hotspot area to different CDN nodes according to the nodes computing
Zhang, et al. Expires 20 April 2023 [Page 7]
Internet-Draft SRv6 Based Computing interconnection Usecases April 2022
load obtained in real-time. Corresponding SRv6 paths are established
for steering different users's packet to different CDN nodes. Through
coordinating computing and network, the problem of unbalanced resource
allocation can be solved and user service experience can be improved.
/-----------------------------------/
/ /
/ Computing and Network Controlle /
/ (CNC) /
/--------:-------:--------:---------/
: : :
: : :
............................: : :..................
+--:----+ : :
+-------+ | : :
|CDN | | : :
|Node 1 |-+ /-------------------------:------------------------/ :
+--+----+ / +----------------+ / :
| / +----------+ | Cloud | +----------+ / :
+--------+--|SDWAN CPE1|--|Specific Network|--|SDWAN CPE3|--+--+ :
/ +----------+ | | +----------+ / | :
/ * +----------------+ * / | :
/ * | * / | :
/ * +----------+ * / +----+-:+
/ **SRv6 path1**|SDWAN CPE2|**SRv6 path2** /+-------+ |
/ +----------+ / |CDN | |
/ | / |Node 2 |-+
/--------------------------------+-----------------/ +-------+
|
|-------------|-----+-----|------------|
+---+--+ +---+--+ +--+---+ +--+---+
+------+ | +------+ | +------+ | +------+ |
|user 1|-+ |user 2|-+ |user 3|-+ |user 4|-+
+------+ +------+ +------+ +------+
Figure 2: CDN system with SRv6 based network
Specifically, as shown in figure 2, CDN Node 1 and CDN Node 2 are
located at SD-WAN CPE1 and SD-WAN CPE3 respectively. There are a large
number of users in the same area accessing to SD-WAN CPE2. It is
assumed that CDN Node 1 is closer to this area. In the traditional way,
user 1 to user 4 all access CDN Node 1. After the computing
interconnection network is deployed, CNC will consider the network and
computing load factors at the same time. User 1 and user 2 will access
CDN Node 1. User 3 and user 4 will access CDN Node 2. Meanwhile, two
SRv6 paths is established on SD-WAN CPE2. Path 1 is to CDN Node 1 and
another is to CDN Node 2. User 1 and user 2 will access through path 1,
and user 3 and user 4 will access another. In this way, better resource
utilization and service experience can be achieved.
Zhang, et al. Expires 20 April 2023 [Page 8]
Internet-Draft SRv6 Based Computing interconnection Usecases April 2022
5. Security Considerations
To be done.
6. IANA Considerations
This document does not make any IANA request.
7. Informative References
[RFC8402] C. Filsfils, S. Previdi, L. Ginsberg, "Segment Routing
Architecture Services", BCP 126, RFC 8402, DOI
10.17487/RFC8402, July 2018,
<https://www.rfc-editor.org/info/rfc8402>.
Authors' Addresses
Xiaoqiu Zhang
China Mobile
Email: zhangxiaoqiu@chinamobile.com
Feng Yang
China Mobile
Email: yangfeng@chinamobile.com
Weiqiang Cheng
China Mobile
Email: chengweiqiang@chinamobile.com
Zhihua Fu
New H3C Technologies
Email: fuzhihua@h3c.com
Zhang, et al. Expires 20 April 2023 [Page 9]
