Page tree

Anuket Project

Skip to end of metadata
Go to start of metadata

This is for Lakelse Release - Issue #2568

8. Hybrid Multi-Cloud: Data Centre to Edge

Table of Contents

8.1 Introduction

The Reference Model Chapter 3 focuses on cloud infrastructure abstractions. While these are generic abstractions they and the associated capabilities of the cloud infrastructure are specified for data centres, central office and colocation centres. The environmental conditions, facility and other constraints, and the variability of deployments on the edge are significantly different and, thus, require separate consideration.

It is unrealistic to expect that a private cloud can cost effectively meet the need of all loads, including peak loads and disaster recovery. It is for that reason that enterprises will implement a hybrid cloud. In a hybrid cloud deployment, at least two or more distinct cloud infrastructures are inter-connected together. In a multi-cloud the distinct cloud infrastructures of the hybrid cloud may be implemented using one or more technologies. The hybrid multi-cloud infrastructure has differences requiring different abstractions. These hybrid multi-clouds can be considered to be federated.

In the Reference Model Chapter 3, the cloud infrastructure is defined. The tenants are required to provide certain needed services (such as Load Balancer (LB), messaging). Thus, the VNF/CNFs incorporate different versions of the same services with the resultant issues related to an explosion of services, their integration and management complexities. To mitigate these issues, the Reference Model must specify the common services that every Telco cloud must support and thereby require workload developers to utilise these pre-specified services.

A generic Telco cloud is a hybrid multi-cloud or a Federated cloud that has deployments in large data centres, central offices or colocation facilities, and the edge. This chapter discusses the characteristics of Telco Edge and hybrid multi-cloud.

8.2 Hybrid Multi-Cloud Architecture

The GSMA whitepaper on "Operator Platform Concept Phase 1: Edge Cloud Computing" (January 2020) states, "Given the wide diversity of use cases that the operators will tasked to address, from healthcare to industrial IoT, it seems logical for operators to create a generic platform that can package the existing assets and capabilities (e.g., voice messaging, IP data services, billing, security, identity management, etc. ...) as well as the new ones that 5G makes available (e.g., Edge cloud, network slicing, etc.) in such a way as to create the necessary flexibility required by this new breed of enterprise customers."

Cloud computing has evolved and matured since 2010 when NIST published its definition of cloud computing, with its 5 essential characteristics, 3 service models and 4 deployment models.

The generic model for an enterprise cloud has to be "hybrid" with the special cases of purely private or public clouds as subsets of the generic hybrid cloud deployment model. In a hybrid cloud deployment, at least two or more distinct cloud infrastructures are inter-connected together.

Cloud deployments can be created using a variety of technologies (e.g., OpenStack, Kubernetes) and commercial technologies (e.g., VMware, AWS, Azure, etc.). A multi-cloud deployment can consist of the use of more than one technology.

A generic Telco cloud is a hybrid multi-cloud. A better designation would be a federation of clouds - a federated cloud:

  • a collection of cooperating, interoperable autonomous component clouds
  • the component clouds perform their local operations (internal requests) while also participating in the federation and responding to other component clouds (external requests)
    • the component clouds are autonomous in terms of, for example, execution autonomy; please note that in a centralised control plane scenario (please see the section "Centralised Control Plane" in the "Edge Computing: Next Steps in Architecture, Design and Testing" whitepaper [26]) the edge clouds do not have total autonomy and are subject to constraints (e.g., workload LCM)
    • execution autonomy is the ability of a component cloud to decide the order in which internal and external requests are performed
  • the component clouds are loosely coupled where no changes are required to participate in a federation
    • also, a federation controller does not impose changes to the component cloud except for running some central component(s) of the federated system (for example, a broker agent – executes as a workload)
  • the component clouds are likely to differ in, for example, infrastructure resources and their cloud platform software
  • workloads may be distributed on single or multiple clouds, where the clouds may be collocated or geographically distributed
  • component clouds only surface NBIs (Please note that VMware deployed in a private and a public cloud can be treated as a single cloud instance)

8.2.1 Characteristics of a Federated Cloud

In this section we will further explore the characteristics of the federated cloud architecture, and architecture building blocks that constitute the federated cloud. For example, Figure 8-1 shows a Telco Cloud that consists of 4 sub-clouds: Private on premise, Cloud Vendor provided on premise, Private outsourced (Commercial Cloud Provider such as a Hyperscaler Cloud Provider (HCP), and Public outsourced (see diagram below). Such an implementation of a Telco Cloud allows for mix'n'match of price points, flexibility in market positioning and time to market, capacity with the objective of attaining near "unlimited" capacity, scaling within a sub-cloud or through bursting across sub-clouds, access to "local" capacity near user base, and access to specialised services.

Example Hybrid Multi-Cloud Component Cloud

Figure 8-1: Example Hybrid Multi-Cloud Component Cloud

8.2.2 Telco Cloud

The Figure 8-2 presents a visualisation of a Telco operator cloud (or simply, Telco cloud) with clouds and cloud components distributed across Regional Data Centres, Metro locations (such as Central Office or a Colocation site) and at the Edge, that are interconnected using a partial mesh network. Please note that at the Regional centre level the interconnections are likely to be a "fuller" mesh while being a sparser mesh at the Edges.

Figure 8-2: Telco Cloud: Data Centre to Edge

The Telco Operator may own and/or have partnerships and network connections to utilize multiple Clouds for network services, IT workloads, and external subscribers. The types of the component clouds include:

  • On Premise Private
    • Open source; Operator or Vendor deployed and managed | OpenStack or Kubernetes based
    • Vendor developed; Operator or Vendor deployed and managed | Examples: Azure on Prem, VMware, Packet, Nokia, Ericsson, etc.
  • On Premise Public: Commercial Cloud service hosted at Operator location but for both Operator and Public use | Example: AWS Wavelength
  • Outsourced Private: hosting outsourced; hosting can be at a Commercial Cloud Service | Examples: Equinix, AWS, etc.
  • (Outsourced) Public: Commercial Cloud Service | Examples: AWS, Azure, VMware, etc.
  • Multiple different Clouds can be co-located in the same physical location and may share some of the physical infrastructure (for example, racks)

In general, a Telco Cloud consists of multiple interconnected very large data centres that serve trans-continental areas (Regions). A Telco Cloud Region may connect to multiple regions of another Telco Cloud via large capacity networks. A Telco Cloud also consists of interconnected local/metro sites (multiple possible scenarios). A local site cloud may connect to multiple Regions within that Telco Cloud or another Telco Cloud. A Telco Cloud also consists of a large number of interconnected edge nodes where these edge nodes maybe impermanent. A Telco Cloud's Edge node may connect to multiple local sites within that Telco Cloud or another Telco Cloud; an Edge node may rarely connect to a Telco Cloud Region.

Table 8-1 captures the essential information about the types of deployments, and responsible parties for cloud artefacts.

TypeSystem DeveloperSystem MaintenanceSystem Operated & Managed byLocation where DeployedPrimary Resource Consumption Models
Private (Internal Users)Open SourceSelf/VendorSelf/VendorOn PremiseReserved, Dedicated
PrivateVendor | HCPSelf/VendorSelf/VendorOn PremiseReserved, Dedicated
PublicVendor | HCPSelf/VendorSelf/VendorOn PremiseReserved, On Demand
PrivateHCPVendorVendorVendor LocationsReserved, Dedicated
Public (All Users)HCPVendorVendorVendor LocationsOn Demand, Reserved

Table 8-1. Cloud Types and the Parties Responsible for Artefacts

8.2.3 Telco Operator Platform Conceptual Architecture

Figure 8-3 shows a conceptual Telco Operator Platform Architecture. The Cloud Infrastructure Resources Layer exposes virtualised (including containerised) resources on the physical infrastructure resources and also consists of various virtualisation and management software (see details later in this chapter). The Cloud Platform Components Layer makes available both elementary and composite objects for use by application and service developers, and for use by Services during runtime. The Cloud Services Layer exposes the Services and Applications that are available to the Users; some of the Services and Applications may be sourced from or execute on other cloud platforms. Please note that while the architecture is shown as a set of layers, this is not an isolation mechanism and, thus, for example, Users may access the Cloud Infrastructure Resources directly without interacting with a Broker.

Conceptual Architecture of a Telco Operator Platform

Figure 8-3: Conceptual Architecture of a Telco Operator Platform

The Cloud Services and the Cloud Resources Brokers provide value-added services in addition to the fundamental capabilities like service and resource discovery. These Brokers are critical for a multi-cloud environment to function and utilise cloud specific plugins to perform the necessary activities. These Brokers can, for example, provision and manage environments with resources and services for Machine Learning (ML) services, Augmented/Virtual Reality, or specific industries.

8.2.4 Multi-Cloud Interactions Model

To realise a federated cloud requires definition and agreement of set of APIs. These APIs should allow each of the parties to interact cooperatively and need to cover the management layer: business management and service operations interactions; as well as the data plane, customer and user, transactions and conversational interfaces.

As outlined in "Figure 8-3 - Conceptual Architecture of a Telco Operator Platform" above the exposure point for the Management Interactions is the "Cloud Service Broker" and the "Cloud Service Broker". The set of interactions that these interface points need to provide are defined by the "Multii-Cloud Interaction Model". This provides a taxonomy for the interactions and is illustrated below:

Figure 8-4: Multi-Cloud Interactions Model

The model defines the following core roles:a

  • Communications Service Provider (CSP) - is the party responsible for providing end user service to their customer
  • Customer / User - are the parties that use the service (User) and establishes the business agreement for the service provision (Customer). For retail services the customer and user are the same party, while for enterprise services the Enterprise is the Customer (responsible for the business agreement) and its representatives are the Users.
  • Cloud Providers - are the parties providing the cloud services. These services could be any XaaS service. It could be that a CSP has an agreement with a SaaS Cloud, which in turn uses an IaaS Cloud Provider to deliver their service.

The set of high level interactions cover:

  • Manage Account - covering Account, Users, Billing
  • Manage Connectivity - Public or Private Network, VPN Configuration, CSP Edge / Cloud Connection Configuration, Connection Security Profile
  • Manage Resource - Resource Pool Management, VM / VNF Management (CPU, Memory, Storage, Network), Image Repository Management, Storage Management, VNF / CNF LCM, Monitor Resources
  • Manage App/VNF - Image / Container / Registry Management, Deploy/Configure/Scale/Start/Stop App/VNF, Monitor App/VNFs
  • Transactions / Conversations - Use Communications Services, Use Edge Applications Services, Use Cloud Services Stereo-Typical Scenarios

A set of stereo-typical interactions cases are illustrated for the cases of a Simple Infrastructure-as-a-Service (IaaS) and Software-as-a-Service (SaaS), where deployment is on a Cloud Provider's centralised sites and/or Edge sites. The scenarios help highlight needs for the Cloud Service Broker and Cloud Resources Broker (as per Figure 8.3) and hence extent of orchestration required to manage the interactions.

Figure 8-4: Simple Stereo-Type Interactions

The following patterns are visible:

For IaaS Cloud Integration:

  • Cloud behaves like VIM and so needs Orchestration
  • Depending on whether the cloud is accessed via public internet or private connection will change the extend of the Connectivity Management

For SaaS Cloud Integration:

  • Cloud behaviors is like subscription management, so avoids need for complex orchestration

For CaaS Cloud Integration:

  • Registry for pulling Containers could be from:
    • Cloud in which case consumption model is closer to SaaS or
    • from Private / Public Registry in which case integration model requires specific registry management elements

For Edge Cloud Integration:

  • Adds need for Communications Service Provider and Cloud Provider physical, network underlay and overlay connectivity management

A disaggregated scenario for a CSP using SaaS who uses IaaS is illustrated in the following diagram:

8.3 Telco Edge Cloud

This section presents the characteristics and capabilities of different Edge cloud deployment locations, infrastructure, footprint, etc. Please note that in the literature many terms are used and, thus, this section includes a table that tries to map these different terms.

8.3.1. Telco Edge Cloud: Deployment Environment Characteristics

Telco Edge Cloud (TEC) deployment locations can be environmentally friendly such as indoors (offices, buildings, etc.) or environmentally challenged such as outdoors (near network radios, curb side, etc.) or environmentally harsh environments (factories, noise, chemical, heat and electromagnetic exposure, etc). Some of the more salient characteristics are captured in Table 8-2.

Facility TypeEnvironmental CharacteristicsCapabilitiesPhysical SecurityImplicationsDeployment Locations
Environmentally friendlyIndoors: typical commercial or residential structuresProtected
Safe for common infrastructure
Easy access to continuous electric power
High/Medium bandwidth Fixed and/or wireless network access
Controlled AccessCommoditised infrastructure with no or minimal need for hardening/ruggedisation
Operational benefits for installation and maintenance
Indoor venues: homes, shops, offices, stationary and secure cabinets
Data centers, central offices, co-location facilities, Vendor premises, Customer premises
Environmentally challengedOutdoors and/or exposed to environmentally harsh conditionsmaybe unprotected
Exposure to abnormal levels of noise, vibration, heat, chemical, electromagnetic pollution
May only have battery power
Low/Medium bandwidth Fixed and/or mobile network access
No or minimal access controlExpensive ruggedisation
Operationally complex
Example locations: curb side, near cellular radios,

Table 8-2. TEC Deployment Location Characteristics & Capabilities

8.3.2 Telco Edge Cloud: Infrastructure Characteristics

Commodity hardware is only suited for environmentally friendly environments. Commodity hardware have standardised designs and form factors. Cloud deployments in data centres typically use such commodity hardware with standardised configurations resulting in operational benefits for procurement, installation and ongoing operations.

In addition to the type of infrastructure hosted in data centre clouds, facilities with smaller sized infrastructure deployments, such as central offices or co-location facilities, may also host non-standard hardware designs including specialised components. The introduction of specialised hardware and custom configurations increases the cloud operations and management complexity.

At the edge, the infrastructure may further include ruggedised hardware for harsh environments and hardware with different form factors.

8.3.3 Telco Edge Cloud: Infrastructure Profiles

The Cloud Infrastructure Profiles section specifies two infrastructure profiles:

The Basic cloud infrastructure profile is intended for use by both IT and Network Function workloads that have low to medium network throughput requirements.

The High Performance cloud infrastructure profile is intended for use by applications that have high network throughput requirements (up to 50Gbps).

The High Performance profile can specify extensions for hardware offloading; please see Hardware Acceleration Abstraction. The Reference Model High Performance profile includes an initial set of High Performance profile extensions.

Based on the infrastructure deployed at the edge, Table 8-3 specifies the Infrastructure Profile features and requirements that would need to be relaxed.

Table 8-3. TEC Exceptions to Infrastructure Profile features and requirements

ReferenceFeatureDescriptionAs Specified in RM Chapter 05
Exception for Edge

Basic TypeHigh PerformanceBasic TypeHigh Performance
infra.stg.cfg.003Storage with replication
infra.stg.cfg.004Storage with encryption
infra.hw.cpu.cfg.001Minimum Number of CPU socketsThis determines the minimum number of CPU sockets within each host2211
infra.hw.cpu.cfg.002Minimum Number of cores per CPUThis determines the number of cores needed per CPU.202011
infra.hw.cpu.cfg.003NUMA alignmentNUMA alignment support and BIOS configured to enable NUMANYNY*

* immaterial if the number of CPU sockets (infra.hw.cpu.cfg.001) is 1

Please note that none of the listed parameters form part of a typical OpenStack flavour except that the vCPU and memory requirements of a flavour cannot exceed the available hardware capacity.

8.3.4 Telco Edge Cloud: Platform Services Deployment

This section characterises the hardware capabilities for different edge deployments and the Platform services that run on the infrastructure. Please note, that the Platform services are containerised to save resources, and benefit from intrinsic availability and auto-scaling capabilities.

Table 8-4. Characteristics of Infrastructure nodes

Platform Services


Network Services

IdentityImagePlacementComputeNetworkingMessage QueueDB Server
EphemeralPersistent BlockPersistent Object
ManagementUnderlay (Provider)Overlay
Control Nodes

Workload Nodes

Storage Nodes

Depending on the facility capabilities, deployments at the edge may be similar to one of the following:

  • Small footprint edge device
  • Single server: deploy multiple (one or more) workloads
  • Single server: single Controller and multiple (one or more) workloads
  • HA at edge (at least 2 edge servers): Multiple Controller and multiple workloads

8.3.5 Comparison of Deployment Topologies and Edge terms

Table 8-5. Comparison of Deployment Topologies

This SpecificationComputeStorageNetworkingRTTSecurityScalabilityElasticityResiliencyPreferred Workload ArchitectureUpgrades
OpenStackOPNFV EdgeEdge GlossaryGSMA
Regional Data Centre (DC)



>1 CPU

>20 cores/CPU
10's EB


HDD and NVMe

>100 Gbps

~100 msHighly SecureHorizontal and unlimited scalingRapid spin up and downInfrastructure architected for resiliency

Redundancy for FT and HA
Microservices based


Hosted on Containers
HW Refresh: ?

Firmware: When required

Platform SW: CD

Central Data Centre

Metro Data Centres
10's to 100's


>1 CPU

>20 cores/CPU
100's PB


NVMe on PCIe

> 100 Gbps

~10 msHighly SecureHorizontal but limited scalingRapid spin up and downInfrastructure architected for some level of resiliency

Redundancy for limited FT and HA
Microservices based


Hosted on Containers
HW Refresh: ?

Firmware: When required

Platform SW: CD

Edge SiteLarge EdgeAggregation Edge
Fixed / Mobile

Some Variability

>=1 CPU

>10 cores/CPU
100 TB


NVMe on PCIe

Permanence / Ephemeral
50 Gbps

~5 msLow Level of TrustHorizontal but highly constrained scaling, if anyRapid spin up (when possible) and downApplications designed for resiliency against infra failures

No or highly limited redundancy
Microservices based


Hosted on Containers
HW Refresh: ?

Firmware: When required

Platform SW: CD

Far Edge SiteMedium EdgeAccess Edge / Aggregation Edge
Mobile / Fixed

High Variability

Harsh Environments


>2 cores/CPU
10's GB



10 Gbps

Connectivity not Guaranteed
<2 ms

Located in network proximity of EUD/IoT
UntrustedLimited Vertical Scaling (resizing)ConstrainedApplications designed for resiliency against infra failures

No or highly limited redundancy
Microservices based or monolithic

Stateless or Stateful

Hosted on Containers or VMs

Subject to QoS, adaptive to resource availability, viz. reduce resource consumption as they saturate
HW Refresh: ?
Firmware: ?

Platform SW: ?

Fog Computing (Mostly deprecated terminology)

Extreme Edge

Far Edge
Small EdgeAccess Edge
  • No labels