Thinking about the cost trade-offs of NFV

Thinking about the cost trade-offs of NFV

We are getting our presentation ready for a webinar on September 12 hosted by IHS Markit “Exposing the Cost Trade-offs of Cloud Native NFV” featuring Michael Howard, IHS Technology Fellow. The webinar will focus on the impact of virtualization on network operators.  Volta’s portion will use two routing uses cases to talk about why virtualization is so important as networks evolve.

Looking at what virtualization did for servers, we see a model that leveraged the increased power of commodity hardware. In servers, the key enabling technology was the x86 processor and in networking, it is switching ASICs. Server virtualization drove much lower cost and greater agility, both of which are critical to service providers.

In networking, we are seeing a shift away from proprietary ASICs built by vendors like Cisco and Juniper, to commercial products from Broadcom, Mellanox, and others. These ASICs allow original design manufacturers (ODMs) to build high-performance white box switches that are priced much lower than legacy routers.

Both servers and white box switches allow you to run workloads on commodity hardware to reduce cost. However, a major difference is that server virtualization uses hypervisors to support multiple workloads as virtual machines on the powerful processors.  In addition, virtualization software provided the ability to setup and teardown virtual workloads quickly – this type of agility is critical to meeting user needs while controlling costs.

How can routing deliver the same cost reduction and agility as server virtualization?

One approach is to run routing software (such as Cisco IOS or Juniper JUNOS) as a virtual machine on a general-purpose x86 based server. This is the idea behind Network Functions Virtualization (NFV), which decouples functions like routing, DNS, firewall, encryption, NAT, and BRAS from the proprietary hardware. These virtualized routers run as a VM on hypervisors, and if more capacity is needed, you can scale out by adding another virtual machine. However, the amount of processing needed on an x86 server can be substantial which makes the server very expensive compared to low-cost white box switches.

Volta approach to the virtual router runs the control plane software the cloud and use a white box switch as the data plane. This completely disaggregates the router. By using the cloud for processing, it makes it possible to scale out faster and less expensively. In addition, you can run multiple virtual routers on a white box switch because the compute-intensive control plane can scale cheaply in the cloud. This delivers the server virtualization benefit of driving down the cost per workload. With virtual routing, you can also move faster, scaling processing up or down quickly as your network needs change. This gives service providers greater flexibility in delivering new services.

We expect that both models will be used.  For example, a VNF router can run on an x86 embedded in a uCPE device because the throughput requirements will be low. Higher port density and/or increased throughput, even on a platform as small as a DCSG router, will favor the cloud-based control plane.

Since the webinar is interactive, we hope you can join us and let us know what you think.