The SMI extends the deployment of Virtual Network Functions (VNF) and Cloud-Native Network Functions (CNFs) to bare metal servers (Cisco UCS-C servers) with the current release. Also, the SMI supports vertically integrated deployment on bare metal servers.

The following are some of the significant features deploying SMI on Bare Metal servers:

• Elimination of VIM-related overhead on Bare Metal servers

• Zero touch deployment for both VNF and CNF based applications

• Automated infrastructure upgrades

• Exposed API for deployment, configuration, and management to enable automation.

• Addresses edge deployment

  • Provides single compute user plane to run at remote sites

• Scales out without any additional overhead

• Ground up API (NETCONF, REST) driven design and architecture

  • All the interfaces are compliant with northbound NFVO (for instance, NSO).

• Simplification and remote management

• Removes shared storage from the architecture

• Single monitoring endpoint for both server and application health

Note	The SMI has the ability to run virtual machines for legacy applications. Currently, it supports only User Plane Function (UPF). Future releases will support legacy (Cisco and partner) virtual applications.

The SMI enables the deployment of Cluster Manager on Bare Metal servers. The following are some of the salient features of SMI Bare Metal architecture:

• Enables all the application containers to run on the bare metal servers with enough resource isolation

• Provides a migration path for SMI on VM to SMI on bare metal

• Automated bring up at the Data Center

• Hardware agnostic architecture

The following figure depicts the high-level SMI Bare Metal Architecture:

With the help of a Bootable ISO, the SMI Cluster Manager boots the Linux Kernel from the base image. This allows compatibility with most of the standard hardware platforms. A customized script downloads and writes the HD image using the Initial RAM File System. Also, the Bootable ISOs smaller size - 23 Mega Bytes (MB) - reduces latency.

The following figure depicts the operations of the Bootable ISO:

SMI operational components and microservices are deployed on VMs. (Refer to SMI VM Quantities and Sizing for details.)

The SMI Cluster Manager (CM) is also deployed as a VM and is used to bootstrap the deployment of other components and applications.

The CM works with the Virtual Infrastructure Manager (VIM) to instantiate the required VMs. In VMware environments, the CM instantiates the virtual machines (VMs) required for the cluster. In OpenStack environments, the CM makes an API call to an orchestrator or Virtual Network Function Manager (VNFM) to instantiate VMs.

The VMs are deployed with a guest OS that is provided with SMI. Once instantiated, the CM provisions the OS, and deploys or provisions the SMI microservices (for example, K8s control plane, K8s etcd, and so on.).

Once all VMs and K8s components are built, the CM can deploy 5G application Ops Centers, which enable NETCONF/RESTCONF interfaces for application configuration and management. All of these actions are API driven and all can be automated and orchestrated.

Scheduling rules such as affinity and anti-affinity help guide K8s for proper node placement, as well as adding node taint and tolerances. Because K8s uses a declarative method of deployment, operators simply need to update the desired number of services and K8s manages scheduling and maintains the correct number of services, even during failure scenarios.

SMI leverages the native resource reservation controls in K8s.

K8s provides a framework to intelligently place pods on the correct server, VM, and/or node, and assign the appropriate system resources, including:

• Service taints, tolerances, affinity, and anti-affinity rules

  • Provides rules for pod placement across available hardware

  • Prevents resource "hotspots" by separating pods with similar resource profiles

  • Provides high availability (HA) by ensuring secondary instances through pod separation

• CPU reservation

  • Allows applications to specify CPUs/CPU requirements (similar to CPU pinning)

  • Prevents negative impacts from context switching, or noisy/grabby neighbors

• Pod quality of service (QoS) definition (e.g. the quality and range of resources available to the Pods)

  • Guaranteed (resource requests = resource limits)

  • Burstable (resource requests > resource limits)

  • Best effort (no resource requests nor limits)

DSCP is implemented at the network level to manage the quality of service and ensure critical traffic is prioritized.

SMI's Common Execution Environment (CEE) provides OAM capabilities for deployed NFs.

The CEE captures information (key metrics) from the NFs in a centralized way for engineers to debug and troubleshoot the overall solution.

There is only one CEE available per K8s cluster, which provides the common set of tools for all deployed NFs. CEE life cycle is independent of NF and it comes equipped with a dedicated Ops Center, which provides the user interface (CLI) and APIs for managing the monitoring tools.

Cisco's cNFs consist of Helm charts (applications and charts) and Docker files (images).

To simplify and establish consistent operations across the various charts and images that comprise each NF, each NF is designed with an Ops Center. Ops Centers provide a common, stable CLI/API for operators to deploy and manage the NF in a holistic way.

SMI provides the following functionality in relation to NF Ops Centers:

• Common NETCONF, RESTCONF, and CLI interfaces, which allows for integration network orchestrators such as Cisco's Network Services Orchestrator (NSO) without need for a custom network element driver (NED)

• A YANG model for the application

• Audit logging and configuration validation

• Lightweight Directory Access Protocol (LDAP) interface directory information services — for example, Active Directory (AD) — to ensure all applications use a common set of user accounts

• Cisco Smart Licensing integration

• Callbacks into the application to execute operational commands

• NETCONF Access Control (NACM) security model

The Service Mesh enabled through SMI connects and manages messages between all pods and services in the cluster. Using this service mesh, traffic is steered within the cluster to finely control which NF components are part of the traffic flow.

Granular controls such as traffic percentage, or application-based traffic characteristics — for example, access point name (APN), subscription permanent identifier (SUPI), or other layer 7 attribute value pairs (AVPs) — are used to control traffic within the cluster. This control enables selective and precise upgrades, such as "canary upgrades". This limits risk and impact when deploying changes in-service and in production. It also affords the ability to selectively drain or decommission NFs.

Besides traffic management applications, the service mesh aids in tracking the flow of traffic between services and nodes, providing security to prevent unauthorized service access and isolating rogue services.

The Common Data Layer (CDL) component enabled through SMI provides the clean separation of stateful (also known as backing services) and stateless services (e.g. application services).

CDL provides services for efficiently managing stateful subscriber and identity information across all deployed Cisco NFs. The CDL is an in-memory database designed specifically for high performance carrier grade requirements and subscriber data. Separating stateful services in this way allows for the stateless application services to be autonomous, lightweight, upgradable, recoverable, and rapidly scalable.

Stateful services must address the availability, consistency, and portability of state. These typically require replication across one or more containers while maintaining state consistency.

As such, CDL redundancy is achieved by local and remote replication of session data. In addition, a background process scans the data store for inconsistencies, stale data, and corruption, and corrects them both locally and remotely.

Table 1 and Table 2 provide SMI VM quantity and sizing recommendations.

NOTE: Individual NFs are deployed as K8s workers through SMI. They each have their own VM recommendations. Refer to the NF documentation for details.

The following table lists the minimum Bare Metal requirements for deploying SMI Cluster Manager.

Note	The Bare Metal requirements listed in the table for deploying SMI Cluster Manager are for reference only. For specific requirements, contact your Cisco account representative.