Enhanced Platform Awareness Workload Placement

In a legacy, chassis-based architecture, network function suppliers have chosen a specific CPU for the network function. CPUs and the CPU cards are connected through a point-to-point, redundant fabric, such as a Dual Star backplane. The bandwidth and latency are guaranteed across the chassis fabric, and there is often a separate management fabric for separation of data and control. In such an architecture, network functions do not have to deal with much variability.

By contrast, datacenter architectures are highly variable and nondeterministic. Virtual machines may be allocated from physical hosts anywhere within the same datacenter, and both the host and the physical links between these hosts can be oversubscribed. The CPU cores within a virtual machine might belong to different sockets on the physical host, leading to cache and memory access issues. This variability can lead to VNFs with completely different performance characteristics, even when they are placed in the same cloud infrastructure.

To ensure deterministic performance, OpenStack Enhanced Platform Awareness (EPA) attributes can increase the efficiency of the network function for high-touch tasks, such as packet forwarding and security. EPA attributes are discovered during the initial allocation of virtual machines from the Virtualized Infrastructure Manager (VIM).

During the VNF instantiation process, the VNF request characteristics are compared to the virtual machine capabilities in order to allocate workload placement across the corresponding VMs. This design supports advanced placement such as:

• Placing high data rate workloads, such as load balancing and bearer plane forwarding, on VMs that support NUMA affinity, hugepage setup, CPU pinning, and PCI pass through or single root I/O virtualization (SR-IOV)

• Placing best-effort workloads, such as statistics gathering or log output, on “vanilla” VMs

• Placing workloads that form part of the same network service (same service chain) in the same switching domain

• Distributing workloads, such as firewalling, DHCP, or other premise-related tasks, to a remote customer premise device

• Providing advanced security capabilities, such as Quick Assist Technology (QAT) crypto assist and Trusted Platform Module