5G-PPP Software Network White Paper 畫重點筆記

From Webscale to Telco, the Cloud Native Journey

https://5g-ppp.eu/white-papers/

這邊是一篇"畫重點"的筆記，關於 5G & Cloud Native ，From 5G-PPP(5G Infrastructure Public Private Partnership) 的白皮書。

1 Introduction

• Software in 5G-PPP so far

  • from “boxes” to “functions”, and from “protocols” to “APIs”

• Cloud impact in general

• Cloud impact in telecom ecosystem in particular

  • covering Service-Oriented Architecture (SOA), Microservices Architecture (MSA) and Service-Based Architecture (SBA), the latter being adopted in next generation CORE [15]
  • telcograde enhancements that should be added in frameworks like Kubernetes, monitoring, stateless design, etc…
  • What are the specific requirements between Webscale and telco players?
  • What gaps do we have in container management eco-system to achieve telco-grade performance?
  • Which segments are critical from latency perspective?
  • What features are required to support micro-services architecture in all the segments?
  • Which factors are difficult to implement in telco (from https://12factor.net)?

• 章節介紹如下：

  • 2 介紹了基於服務的體系結構，12個因素以及網絡，虛擬化，編排等所需的技術。
  • 3 列出了電信公司的要求。
  • 4 描述了電信級的增強功能。
  • 5 簡要概述了技術推動因素，開源和標準。
  • 6 結論。

2 Cloud Native: a Webscale View

SBA(Service-Based Architecture)

3 ways to implement:
either as a network element on a dedicated hardware
as a software instance running on a dedicated hardware
as a virtualized / cloud-native function instantiated on an appropriate platform

利用雲軟件技術，3GPP NF架構可實現更高的靈活性，可編程性，自動化以及顯著的成本/能耗降低。

2.1 Microservice Architecture

提供超可靠，低延遲的通信和高速
為此，需要在架構中實現高度靈活性，可歸納如下：

• 支持對網絡的不同要求，如網絡容量，數據速率，傳輸延遲，信息安全;
• 支持向運營商的服務和第三方應用程序公開網絡功能
• 支持輕鬆部署和維護。 每個功能都應該能夠根據新的要求進行升級，並根據系統能力進行橫向擴展，而不會影響其他功能。

SBA benefits to 5G:

• Easy update of Network:
  • Finer granularity allows individual services to be upgraded with minimal impact to other services.
  • Facilitated continuous integration reduces the time-to-market for installing bug fixes and rolling out new network features and operator applications.
• Extensibility:
  • Light-weighted service-based interfaces are needed to communicate across services.
• Modularity & Reusability:
  • The network is composed of modularized services reflecting the network capabilities and provides support to key 5G features such as network slicing.
  • A service can be easily invoked by other services (with appropriate authorization), enabling each service to be reused as much as possible.
• Openness
  • Together with some management and control functions (i.e. authentication, authorization, accounting), the information about a 5G network can be easily exposed to external users such as 3rd parties through a specific service without complicated protocol conversion

2.2 Twelve Factors

https://12factor.net/zh_cn/

2.3 Technology for Cloud Native

Cloud Native Trail Map
https://github.com/cncf/landscape/blob/master/README.md#trail-map

3 Telco Requirements

• 5G Service Enablers
  • eMBB – Enhanced Mobile Broadband
  • mMTC – Massive Machine Type Communications
  • URLLC –Ultra-Reliable and Low Latency Communications

IMT-2020 5G Triangle

• 3.1 Service Requirement: 99.999%
• 3.2 Application Type: like multi-access edge computing (MEC)
• 3.3 Deployment Environment
• 3.4 Operability: Need monitoring system
• 3.5 Business Environment
• 3.6 Human Factors

4 Toward Telco Grade Webscale Frameworks

4.1 Need for Telco Grade features

• Open and modular software

• Intelligence in software

• Scalable and efficient

• Typically, telco key performance indicators taken into account include latency and bandwidth, availability and security.

  • For instance, since each microservice contributes in the overall latency, it is challenging to predict the latency of a specific service, especially in cases of large and distributed systems.
  • container management technologies (Kubernetes, Mesos, Swarm) are using an overlay network (Layer 4) which is not optimized for low latency requirements, e.g. in the cloud RAN
  • network availability, as 99.999% availability is commonly set by the operators. Currently, cloud-native applications are not built to handle bandwidth, latency and network availability requirements.
  • container management include the lack of a dynamic carrier-grade orchestration, the management of containers across different sites and data centers and the lack of centralized controllers for networking aspects with respect to provisioning, resiliency, management/coordination, automatic network configuration and visibility of traffic from container-to-container across networks -whether physical or virtual.

• 為了滿足上述要求，必須滿足以下條件：

  • Automation of operations, upgrade, backup and restore as well as restart capabilities
  • Monitoring, performance counter, alarms and logging/trace support
  • Security and tenant isolation.

4.2 Twelve Factors in Telco Domain

• Build/Release/Run (12.V) is not a part of ETSI standards
• dependency declaration and management (12.II) should be of particular concern
• the codebase (12.I) and Dev/Prod parity (12.X) which should accommodate various deployments strategies over a potentially multi-cluster multi-vendor environment.
• factors 12.VIII (Concurrency) and 12.VI (Processes) 12.IX (Disposability) and 12.XI (Logs) pose the greatest challenges ahead for the participating 5G projects.
• Logs (12.XI) speed, heterogeneity and variability of origin will force a readjustment of current telco methodologies.
• Disposability (12.IX), i.e. small start-up times and graceful termination, is major concern for five nines availability systems, especially for emergency services and in relation to QoS constraints.
• the factor on administrative processes (12.XII) seems a key aspect, and especially critical since it has less to do with technology evolution.

4.3 Telco-Grade Features

Several features must be integrated to the aforementioned frameworks:

1. Multi-network interfaces
2. Service function chaining
3. Data plane acceleration
   • low packet processing time to reach very low latency
   • predictability and stability of packet processing delay to minimize jitter
   • guaranteed and prioritized bandwidth for each Telco workload to prevent network contention
   • low CPU consumption for network tasks to allow low energy usage
   • native network performance for Telco workloads
4. Enhanced platform awareness
   • CPU pinning to avoid unpredictable latency and host CPU overcommit by dedicating CPUs to the Telco containers
   • Numa awareness to improve the utilization of compute resources for Telco containers that need to avoid cross NUMA node memory access
   • huge pages to accelerate the memory management by using larger page sizes.
5. Environment aware scheduling

5 Technology Enablers – Open source and Standards

5.1 Standards

• ETSI NFV https://www.etsi.org/technologies/nfv
• TM Forum https://www.tmforum.org/virtualization/
• Open Networking Foundation Open Networking Foundation (ONF) Central Office Re-architected as a Datacenter (CORD)

5.2 Open Source Software

可以參考 CNCF 的軟體來實作
MANO frameworks such as Linux Foundation’s ONAP [41], ONF XOS [42], Cloudify [43] and ETSI OSM [44] are starting to evolve towards this approach for microservices.

6 Conclusion

• The VNFs are now broken into pieces, support continuous integration and continuous development (CI/CD) and interact using service meshes.
• Compare the three architectures inspired from the domain driven design
  • Service Oriented Architecture
  • Service Based Architecture
  • Microservice Architecture
• the new VNFs, microservice, are designed, a set of technology used to deploy, orchestrate, network and monitor them.
• Telco grade features
• The telco requirements that are needed to be supported in community developed open source.