2010 Prediction: Niel Viljoen, Netronome

Virtualization in 2010 and Beyond

Virtualization is at the core of the convergence of networking and computing. This trend is going to become even more pronounced in 2010 with customers requiring more and more network features from their virtualized servers. Initially the “switches” were very simple Ethernet bridges between the virtual machines, and between the virtual machines and the physical network interface cards. Today, we see the leading edge virtualization environments deploying virtual switches that support access control lists (ACLs) and VLAN support as features.

However, the switch is only one element of the “network” that is found inside the server today, and 2010 will lead us to announcements of even more networking functions inside the next generation data center including firewalls, intrusion prevention, and application optimization – all interconnected inside the virtualized server platform.

The impact that these software features have on network and server performance inside the virtual server is significant.  As the number of virtual machines and cores inside the servers grow this problem becomes more and more pronounced.

The implication is that in the same way that hardware network platforms evolved in the 1980’s and 90’s (the first Cisco routers were workstations!), we will see the same effect on virtualized servers.

Hardware support for switching and the other network-based security functions will start appearing in 2010. This will allow the virtualized server to scale.

A further related trend is the growth of large clusters to support cloud computing using virtualized servers. Traditionally, these servers have been low cost with rudimentary networking; however, in many of the clouds the so called bi-section bandwidth is becoming a significant limitation in the performance of the clouds when running cluster tools such as map reduce implementations. The solution is that greater connectivity needs to be deployed between virtualized servers. New techniques such as so called “fat tree” networks in combination with the embedded switches within the virtualized servers will have a major impact and in my view become the predominant cloud data center design methodology with the core benefits being better performance and significantly lower cost.

So in summary: The network will invade the virtualized server with all its constituent elements and through its internal intelligent switching ability be able to make use of complex routing topologies such as the “fat tree” networks leading to lower cost, higher performing enterprise and public clouds!