Today many look to SDN as the next big revolution in Networking. But why is there such hype? What radical change in the economics of networking will shift the industry? The answer is Virtualization.
Virtualization’s growth is still in its infancy, and many aspects remain unexplored. Still there are aspects of which we are certain:
- With an explosion in the number of Virtual devices, it is unaffordable for humans to remain in the loop for routine network operations.
- Emerging business models are not achievable when (slow) humans are involved in the provisioning process.
Historically proven solutions to these categories of speed and cost challenges exist. If these challenges can be met while at the same time introducing a proper layering of operations APIs, there is an opportunity to enable frictionless configuration of devices and networks. In other words, if rapid configuration and reconfiguration were extensible, new services could be introduced at low incremental cost.
Displacing brittle OSS
The modern computer Operating System (OS) contains layered, abstracted APIs. This is what provides the flexibility for OS subsystems to evolve independently. And with this independence, the OS is economically positioned to serve in diverse environments.
Contrast the modern computer OS with the typical network operator’s Operations Support System (OSS) infrastructure. Over the years network operators have focused on minimizing costs and TTM by incrementally bolting on only those capabilities needed for the latest service. At the same time similar improvised OSS capabilities have been spun-up in parallel by competitors serving the same market segment. The natural result of this pipeline of replicated customization has been the growth of brittle management systems. There are many downsides to brittle systems – one example is that purchasing decisions become driven by the imperative of not upsetting this brittle OSS infrastructure. Sometimes this is the right choice. But if a custom network experience is not the ultimate differentiator for a business, then large segments of a network and its administration become vulnerable to alternative options.
Amazon Web Services (AWS) is one proof point that segments of networking and compute can be outsourced. AWS does several things well: it provides VM capacity in hours instead of weeks, it removes much complexity from the customer interface, and it is partner extensible. In-house compute providers have been forced to match or adopt these outsourced capabilities. There is no reason to assume IT outsourcing will end with Cloud applications like AWS. How long will it take virtual network configuration and operation to be impacted as well?
Cisco Prime is one proof point that real time configuration of virtual networks will happen. As these systems evolve, simplicity argues that these configuration systems will not expose their underlying southbound interfaces (such as onePK) to applications and human operators. It is far simpler to present higher layer network abstractions in order to hide much of the complexity under the covers of implementation.
For developers and partners creating these abstracted models, the question then becomes what complexity should be hidden? Decades of computer OS development experience tells us to layer these interface abstractions, while at the same time providing strong incentives for systems integrators to use the highest level of abstraction practical. Higher layers of abstraction will enable the hiding of existing control protocols, the insertion of new control protocols, the leveraging of existing OSS interfaces, and the introduction new OSS capabilities.
This is a key promise of SDN: leveraging the business changes driven by virtualization so that network intelligence can be based upon higher level abstractions while at the same time simplifying management interfaces away from brittle low level constructs that were designed assuming human involvement.
Industry churn prior to OSS uniformity
It will take years for the winning layering of network abstractions to be selected by various market segments. In some cases a layer of abstraction will be defined by a defacto by market winner, in other cases a layered abstraction will be determined by explicit standardization. It is likely but not inevitable that lower level abstractions will solidify first. In all cases there will be much complexity (and development cost) hidden beneath each implementation delivering a layered abstraction.
A consequence of hiding code beneath layers is that the hidden code cannot affordably be customized except for the largest of operators. OSS infrastructures will become far more uniform than currently seen. There will be benefits and costs to a future of industry wide OSS products:
- Underlying OSS development costs will be spread over large base of customers
- With broader applicability of a particular OSS, dollars invested in those OSSs can be focused on new function delivery rather than the replication of similar but custom capabilities.
- Defacto interfaces to 3rd party virtual application providers
There are consequences in a move to an industry consistent Northbound however. Specifically we should expect less variation in deployment configurations. Standardized configurations are far easier to integrate when outside parties are providing the OSS.
There are many precedents for such dramatic IT shifts from unfettered variation to conformity. Two examples:
- Printer drivers moved from applications and became embedded within the OS. The market found this far more efficient than having such basic capabilities rewritten by every application. Applications were freed to focus on their specific differentiators.
- Salesforce has become successful in part because multiple companies adapted their non-differentiating business processes to the Salesforce OSS.
If virtualization and SDN moves network operations in an industry unified direction, networks will become more homogeneous in their outward appearance, and networks will be exposed via high level intent based GUIs and programming environments. Under the covers will be APIs such as onePK to allow programmatic fine grained control of devices. These changes will be welcomed first by customers for whom the network isn’t their core differentiator. And network configuration and management will move towards being frictionless as configuration speeds drop, and operations costs shrink.