As hot as software-defined networking is today from a media perspective, there are still relatively few production deployments of it from companies not named Google and Baidu. The biggest challenge for mainstream enterprises is trying to understand what, exactly, one would do with an SDN. SDNs promise to deliver a more flexible, agile, programmable, virtual, lower-cost and easier-to-manage network, making it seem like a no-brainer. However, realistically speaking, achieving all of the above makes no sense. Instead of trying to go elephant hunting and solve all of the world’s networking problems and the Middle East crisis all at once, IT leaders should take a step back and look for use-cases that create small victories. Normally, I would call this low-hanging fruit, but I don’t really believe any of this is low-hanging fruit right now, so let’s focus on fruit that’s in the middle of the tree but still reachable.
With that being said, here are what I think are the best use cases for SDNs today:
- Rapid provisioning of network services. Without a doubt, this is the closest thing the industry has to low-hanging fruit. With legacy infrastructure, deploying new network services often means deploying new boxes that chain to other boxes, creating a messy network to manage. If the service is located in a branch, the process of provisioning a service can be highly labor-intensive and can have long lead times. SDNs allow for services to be turned on with a few mouse clicks from a centralized console. Over time, I would expect to see more cloud providers, such as Pertino Networks, offer this as a service (NFVaaS?).
- Improved network management and analytics. This certainly isn’t for every company, but for those of you that use analytic tools such as Netscout and Riverbed/Cascade, more network data is generally better. Many of the SDN vendors have rolled out network TAPs as an SDN “application”. While these don’t have near the level of a TAP from a company like Gigamon, they do provide more traffic information from more places and can improve the value of the analytic engine.
- Network virtualization. This is known by some as “campus slicing” or overlay networks, and can be used to create isolated virtual networks from a single physical network — much like server virtualization does with compute. This is best-suited to organizations that currently run multiple physical networks and are trying to consolidate down and have more of a shared services model. This is most common with government organizations, as well as large conglomerates.
- Network programmability. This is an interesting use case for SDN, in that it’s really not a new concept. F5 has had iRules for years, and Extreme has had a programming interface into its operating system, XOS. The difference today is that more network managers understand the concept of it and are becoming more comfortable with it. This is a bit higher on the fruit tree but, over time, should allow companies to customize the network and rapidly add new features.
When it comes to SDN use cases there are really no ‘right answers’, as it really depends on the goals of the network team. All things being equal, I would recommend looking first at virtualizing network services, as this seems to be the most manually intensive and costly process today.