Alistair Croll has a great post on GIGAOM discussing how networking vendors will need to change in order to support a cloud computing infrastructure.

He outlines two options for networking vendors that will keep them relevant in a cloud computing environment.

In option number one he postulates that virtual appliances are the way to go, that the "pendulum swings back to software". Option number two revolves around sales strategy, and he suggests that networking vendors will need to sell to the providers of the cloud. That makes sense to me. If you want to be a part of the cloud computing infrastructure then you should probably sell to the people building those infrastructures. Option #2 is not really a question of technology; Alistair is somewhat mixing his metaphors, as it were, with these two options, so I'm going to focus on the technology because honestly, I couldn't sell water to a man dying of thirst.

The goal of the virtual appliance (option #1) is to provide networking (routing, firewalls, load balancing) to customers of the cloud in a way that is (a) easy for cloud computing providers to provision and de-provision on-demand and (b) segregates management and configuration on a customer by customer basis. When Customer A wants a load balancer, for example, the cloud computing provider needs to provision the resource and associated management to the customer in an automated way. Hence Alistair's choice of a virtual appliance, as it is well established that within the right framework, with the right tools, a virtual image of any application can be easily provisioned and de-provisioned.

While the virtual appliance is certainly one way in which this can be accomplished, there is another option that doesn't introduce the overhead of virtualization into the equation and maintains the performance gains made in hardware in the networking world.

The cloud computing environment must be scalable on-demand, and it must support high volume (capacity) without sacrificing performance. Automation and integration have always, always, degraded performance in the software world and it is rarely as "on-demand" as the solutions claim to be. While a virtual appliance is not necessarily a bad idea, there are still questions regarding the impact of virtualization on performance of networking devices that have not been answered. The loss of acceleration hardware assistance for compute intensive tasks like XML processing, compression, SSL operations, bulk encryption, and simple packet processing at wire speed might unnecessarily increase the hardware horsepower needed to deploy such solutions, and it is still unlikely to come close to the performance and capacity available in hardware solutions.

For highly performant cloud computing infrastructures, option #3 is likely to be a better choice because it offers flexibility, integration capabilities, virtualization of management, and automation of provisioning (on-demand) without sacrificing the hard won benefits of hardware processing in the network.