…….because really it’s the severity of the pain point that will make you consider it or not
During a recent meeting, a well-known cloud player challenged the idea of SDN. They believed the technology was not evolved enough and would “gain little” from its deployment.
What was their SDN rationale? The customer maintains several business units driving massive scaling needs for on-demand compute and storage. They want their network to be invisible, never impacting the application environment. Clearly, their data centers are not a place where they would take chances. At least, chances in their minds.
Not surprisingly, their pain points were a familiar refrain for their type of organization:
- Deploying new hardware, such as moving from 1 Gig to 10 Gig, and selecting new vendors caused the consistent need for retraining, management and operational framework changes. These disruptions are frequently lead to some customers adopting a one-vendor, end-to-end approach.
- Higher OpEx than CapEx to the point where 90% of costs are operational. Physical devices continue to drop in price, due to memory costs dropping and through higher degrees of system integration. While some call this the commoditization of hardware, capital expenses are no longer dominating the IT budget.
- Maintaining an inherently inflexible and expensive-to-change topology. Networks have been built to be solid, stable and that means not making them easy to change. With the onset of cloud, workload management now requires varying degrees of faster network provisioning, changing the network configuration to accommodate a change in the workload environment.
Ironically, I would bet most SDN advocates would say “Hang on, those are benefits SDN should deliver.” And indeed, I am one of those advocates
If you have open SDN:
- You will be abstracting the network operating system from the hardware, and therefore abstract applications from the network. This is common thinking with white box servers, and mirroring this level of abstraction in networking is a key step. Once you have this freedom to leverage a consistent software environment, hardware changes do not drive a ripple of management and operational issues.
- If your hardware is abstracted, and commoditized, you can now think of new topologies that scale by adding common building blocks both to add horizontal scale– the number of racks or servers, and vertical scale, to enable aggregation of massive arrays of servers to have a line rate or near line rate path out of the data center. Building out cloud / compute resources with racks, rows and data centers full of CPUs is indeed the parallel thought that is now widely adopted.
- You will be able to program the network, without making physical topology changes, again helping to reduce OpEx and unplanned downtime. Programmability based on ‘’gold standard’’ policies and scripts is again a common means to automate aspects of server management. Networking can move in this direction through the idea of external programming will help IT teams tackle operation pain — both costs and unplanned ‘’fat finger’’ downtime.
Whether you call these ideas SDN or open networking, it does not matter. The idea is there are real problems today that are addressable today. All companies with large data center properties need to take a look at how SDN can help them today, there is no reason to wait!