As the Verizon Wireless OpenStack Edge platform grows, making efficient use of resources also grows in importance, to deliver the cost benefits of OpenStack to Verizon customers. This presentation explores how utilization at the highest level, Virtual Network Function (VNF) redundancy pairs, relates to utilizations at intermediate and lower levels. For example, an intermediate level, a nova server, corresponds to a lower level Libvirt instance, which in turn is composed of CPU cores which each have utilizations. In some cases, these CPU cores may be pinned by the VNF design, and used at 100%, in which case no optimization is possible. In other cases, some over-subscription may be possible without impacting services. In addition to tying a given nova VM together with its foundational levels, each VNF frequently contains multiple nova instances, neutron networks, and cinder volumes, which also must be considered when over-subscribing.
Attendees should expect to learn how Verizon Wireless maps OpenStack services for a given Virtualized Network Function (VNF) to each other, underlying services and hardware, and Virtualized Network Functions (VNFs). This tying together has benefits both for efficiency, via over-subscription, and reliability, via rapid fault isolation. For a given VNF, many OpenStack services are involved, typically including nova instances, neutron networks, cinder volumes, and more. Each of these services is then supported by underlying layers; in the case of nova, KVM and CPU cores, which are in turn supported by computes, which are in turn correlated with hardware data. As some CPU cores are often pinned in VNFs, and may be utilized at 100%, over-subscription is not always possible. The VNF site-redundant pair is also monitored at a high level to ensure that any over-subscription will not adversely impact key functionality.