Uredi

Deli z drugimi prek


Near-edge compute

Azure Operator Nexus offers a group of on-premises cloud solutions. One of the on-premises offerings allows telco operators to run the network functions in a near-edge environment.

In a near-edge environment (also known as an instance), the compute servers (also known as bare-metal machines) represent the physical machines on the rack. They run the Azure Linux operating system and provide support for running high-performance workloads.

Compute connectivity

This diagram shows the connectivity model followed by computes in the near-edge instances.

Diagram of Azure Operator Nexus compute connectivity.

Compute configurations

Azure Operator Nexus supports a range of geometries and configurations. This table specifies the resources available per compute.

Property Specification/Description
Number of virtual CPUs (vCPUs) for tenant usage 96 vCPUs, with hyperthreading enabled per compute server.
Number of vCPUs available for workloads 2 to 48 vCPUs, with an even number of vCPUs only. No cross-NUMA (nonuniform memory access) virtual machines (VMs).
CPU pinning Default.
RAM for running tenant workloads 448 GiB (224 GiB per NUMA).
Huge pages for tenant workloads All VMs are backed by 1GiB (1G) huge pages.
Disk (ephemeral) per compute Up to 3.5 TB (3.18 TiB) per compute host.
Data plane traffic path for workloads Single-root I/O virtualization (SR-IOV).
Number of SR-IOV virtual functions (VFs) Maximum of 32 virtual NICs (vNICs), with 30 VFs available for tenant workloads per NUMA.
SR-IOV NIC support Enabled on all 100G NIC ports on VMs, with VFs assigned out of Mellanox-supported VF link aggregation (VF LAG). The allocated VFs are from the same physical NIC and within the same NUMA boundary. NIC ports that provide VF LAG are connected to two different top-of-rack (ToR) switches for redundancy.

Support includes trunked VF receive-side scaling (RSS) with hardware queuing. Support also includes multiple queues on VMs.
IPv4/IPv6 support Dual-stack IPv4/IPv6, IPv4, and IPv6-only virtual machines.