Internet-Draft | SFC Scheduling Algorithm | December 2022 |
Wu, et al. | Expires 26 June 2023 | [Page] |
This document provides the application examples of mapping and deployment algorithms to address the problems of large resource overhead and end-to-end latency in Service Function Chain(SFC) that cannot meet the requirements of latency-sensitive applications in terms of both latency optimization and resource optimization.In terms of delay-optimized mapping and deployment of SFC, the application example of granularity-variable SFC mapping algorithm based on microservice architecture reduces the number of instantiated microservice units and the number of end-to-end routing hops by merging redundant microservice units within the service function chain and reusing microservice units between the service function chains.In terms of resource-optimized mapping and deployment of SFC, the application example of SFC mapping algorithm based on improved gray wolf optimization algorithm optimizes the mapping of SFC through two strategies of reverse learning initialization and nonlinear convergence factor, and improves the efficiency of the mapping scheme.¶
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 26 June 2023.¶
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.¶
In Network Function Virtualization (NFV) environment, user's service is done by Service Function Chain (SFC) composed by Virtual Network Function (VNF) in a certain order, so user's request can be called SFC Request ( SFC Request, SFCR). Given a series of SFCRs, it requires instantiating and deploying the corresponding VNFs to complete these services. The software features of VNFs allow more flexibility in the selection of deployment locations and resource allocation. In addition, during the process of SFCR mapping, it is necessary to consider the sequential constraints of the VNFs that make up its SFCs to avoid additional bandwidth usage as much as possible. Thus, in the process of SFC scheduling and VNF deployment, it is necessary to design an optimal solution to improve the resource utilization of the network, which is also known as the SFC scheduling and VNF deployment problem.¶
The scheduling of service function chains and deployment of virtual network functions include three main parts: determination of the mapping scheme, deployment and management, among which the determination of the mapping scheme as a key step in the scheduling will affect the costs associated with the subsequent processes. How to reasonably deploy each virtual network function which constitutes the service function chain in the physical topology and reduce the overhead of physical resources is a major challenge for the current mapping scheme determination problem. In addition, with the development of emerging technologies such as 5G, many latency-sensitive applications such as virtual reality, augmented reality, and autonomous driving have emerged, which put forward high requirements on the end-to-end latency of Service Function Chain, and how to reduce the end-to-end latency of Service Function Chain to a certain extent is another major optimization goal and challenge of the mapping problem.¶
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT","SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119 [RFC2119] RFC 8174 [RFC8174].¶
This document uses the following terms from RFC 7665 [RFC7665] RFC 8763 [RFC8763]:¶
This document introduces the following terms:¶
Network resources in network topology are limited, and Network Function Virtualization (NFV) presents new challenges in terms of how to efficiently map and deploy Service Function Chain (SFC) based on user's Service Function Chain Request (SFCR) while saving operators' cost. This document presents examples of Variable Granularity Service Function Chain Mapping(VG-SFCM) based on microservices architecture algorithm and Improved Grey Wolf Optimization Service Function Chain Mapping ( IMGWO-SFCM ) algorithm for both latency optimization and resource optimization respectively.¶
There are many logical functions that are repeatedly performed between different individual VNFs within a single SFC, such as packet input and output, message classification, etc. In the process of SFC mapping, considering the minimum granularity of the mapping and performing microservice splitting and merging can effectively remove these redundant logical functions and reduce the end-to-end delay and resource overhead incurred during deployment according to the mapping solution. Considering the splitting and optimization of VNFs within SFCs, we propose a Variable Granularity Service Function Chain Mapping algorithm (VG-SFCM) based on microservice architecture.¶
Firstly, the set of microservice units (AtomNFs) constituting the original SFC is obtained by fast splitting based on typical VNF, and then the redundant AtomNFs within a single SFC are merged by applying the merging strategy of redundant AtomNFs and drawn to form a new SFC service graph, then finally, the instantiation of AtomNFs is further reduced by the AtomNF reuse strategy among SFCs.¶
The splitting of VNFs is mainly based on the set of VNFs and their corresponding logical functions, and the VNFs are split into several fine-grained AtomNFs, which cannot be fully automated at this stage due to the different functions provided by different VNFs, and still requires a lot of manual work to complete the logical analysis and AtomNF definition. In order to realize the fast splitting of SFC and reduce the resource overhead caused by the splitting, the fast splitting based on typical VNF prepares the logical analysis and splitting work in advance by logically analyzing the VNFs that appear more frequently and preparing the AtomNF image file corresponding to the high-frequency VNFs in advance and putting them into the repository. When a SFCR arrives, it can quickly find and directly pull the relevant mirror according to the relevant VNF splitting scheme. If there is a VNF in SFC that is not pre-split, we do not split it and treat it as a coarse-grained single unit.¶
Different VNF may get several identical AtomNF after splitting, and repeated execution of the same AtomNF does not bring actual changes to the data flow instead it increases the resource overhead and end-to-end delay of the corresponding mapping nodes, so we can analyze and merge these multi-occurring AtomNF. Referring to the classification rules of MicroNF, AtomNF is classified into six categories: endpoint, classification, modification, reconfiguration, flow control and static for merging. In order to ensure that the new SFC after merging can provide the same services as before merging, while minimizing the end-to-end latency, the merging of AtomNF needs to comply with the following principles.¶
The reuse of VNF or AtomNF instances can currently be realized by multi-tenancy technology, which enables a single software instance to provide corresponding services for multiple tenants. Compared with the traditional non-reusable mapping scheme, the node resource utilization can be effectively improved by providing services for SFCRs through one reusable AtomNF instance in SFC mapping process. In order to realize the reuse of subsequent requests, the first arriving SFCR need to determine the best mapping scheme through IMGWO-SFCM. After the subsequent arriving SFCRs complete the initial scheduling and splitting and merging work, they judge the intersection of their AtomNF sets with the AtomNF sets of the previous SFCR, and the resulting intersection is the reusable AtomNFs, and then perform the reusability judgment of the relevant microservice unit mapping nodes.After the screening of reusable AtomNFs is completed, the corresponding node is checked for resources to determine whether the remaining resources of the node can carry the remaining non-reusable AtomNFs of the SFC. If it cannot, the relevant AtomNFs are deployed to the neighboring nodes with sufficient resources. When there is no available resource in the neighboring nodes to meet the relevant demand, the mapping scheme is determined by IMGWO-SFCM for the SFC again.¶
Using splitting, merging and multiplexing strategies for SFC, VG-SFCM reduces the instantiation of AtomNF, and effectively reduces ends-to-ends latency of SFC, which can also minimize the cost of network deployment to a certain extent.¶
For SFCR initiated by users at the application layer, the control layer needs to orchestrate it into SFC after receiving it. The SFC deployed according to different mapping schemes consume different computing resources, memory resources and bandwidth resources, and also incur different node activation costs and deployment costs.¶
The mapping and deployment of SFC needs to obey the following four basic principles.¶
First, we search the first K paths between the source and destination nodes by the loop-free KSP search algorithm and search the possible mapping schemes according to the first K shortest paths and the number of VNFs to be deployed in the SFC. Subsequently, the matching of individual gray wolves with the mapping scheme is completed by mapping scheme coding. Then, we initialize the gray wolf population according to the backward learning strategy of Improved Gray Wolf Optimization algorithm (IMGWO), select the dominant gray wolf individuals according to the optimization function, i.e., the fitness function, and then enhance the global search and local search ability of the algorithm in the early stage by the nonlinear convergence strategy of the convergence factor. Finally, we determine the final mapping scheme by iteration.¶
The standard gray wolf optimization algorithm suffers from slow late convergence and easily falls into local optimum, so an improved gray wolf optimization algorithm (IMGWO) is proposed by improving the standard gray wolf optimization algorithm through the group initialization strategy based on reverse learning and the convergence factor nonlinear optimization strategy. The gray wolf optimization group initialization based on the backward learning strategy first randomly generates the initial wolf group and generates the reverse wolf group based on the initial wolf group. Subsequently, the top three individuals with the highest fitness in the two groups are determined according to the fitness function. Meanwhile, in order to improve the global search ability and local search ability of the standard GWO algorithm and avoid falling into local optimal solutions, the nonlinear optimization strategy is applied to the convergence factor a.¶
The group initialization strategy based on backward learning and the convergence factor nonlinear optimization strategy make IMGWO algorithm a good balance of global search and local search ability, which can accelerate the overall convergence speed of the algorithm.¶
The IMGWO algorithm can be divided into seven steps as follows.¶
The steps of the implementation of IMGWO-SFCM algorithm are as follows.¶
Thanks to Hong Chen and Zhang Junnan for the first draft.¶