Lesson 2 of 5: Embrace commodity hardware

Lesson 2 of 5: Embrace commodity hardware

x86 processors from Intel and AMD were used by a range of server ODMs. Moore’s Law held and these chips grew in raw processing power. It meant that the need for custom silicon greatly diminished. Custom silicon was common – think of Sun servers running Sun’s own SPARC processors.

Commodity should not be viewed as a pejorative. Rather, it refers to a product that is widely available as a mass produced, standard commercial product. By using mass produced silicon, server manufacturers could deliver a high performance product at a very attractive price. As volumes increased, economies of scale continued to drive costs lower.

In networking, we are seeing a shift away from proprietary ASICs built by networking vendors like Cisco and Juniper to commodity products from Broadcom, Mellanox, Barefoot and others. These commercial ASICs make it possible for a range of ODMs to build high performance switches that are priced much lower than legacy products. Many legacy vendors already use these switching ASICs. The broader ecosystem has developed in parallel. For example, Open Network Linux (ONL) is broadly supported as an OS for these switches.

Commodity hardware allows operators to reduce costs significantly. That was the real driver behind server virtualization. As commodity hardware became more powerful, multiple workloads could run on a single server thanks to the hypervisor. However, it wasn’t just about the cost of the server. Data centers saw savings in power consumption and reduced space. Of course, that was on a per application workload basis. We know that the lower cost of computing allowed the number of workloads to increase. Service providers can benefit from the same bundle of savings. As the edge of the network transforms because of initiatives like CORD and MEC, the constellation of capital and operational savings becomes critically important.