BACKGROUND IMAGE: iSTOCK/GETTY IMAGES
While the potential benefits of running IT services in the cloud are clear, there are still important challenges to overcome before the full promise of elastic clouds can be realized. The nature of these challenges varies -- affecting cloud networks, compute resources and storage -- and perhaps the most critical place they must be addressed is in hybrid cloud deployments.
In a hybrid cloud setup, one part of a service is run in a customer's facilities, while the remaining part is run in a service provider's cloud. The workload is split on a particular dimension, and enterprises typically determine where an element of the service runs based on performance, availability, privacy, cost and other considerations. To date, however, many customers have held back from committing to hybrid cloud implementations because of concerns about whether the deployment model is capable of meeting their needs.
To highlight an example of the concerns that need to be addressed, consider an analysis researchers at Microsoft did to evaluate performance of an application that was theoretically well-suited for use in hybrid cloud architectures: a distributed MapReduce-based application. The application performed three times faster in a private data center than when running split between the data center and a public cloud computing service.
If providers want to attract customers interested in hybrid cloud deployments, they need to address the limitations of their existing environments. There are several elements of a public cloud that, if enhanced, help improve hybrid cloud support. Candidates for upgrades include virtual computing performance, orchestration system functionality and security assurance capabilities. And for providers that own and operate those networks, the network that connects the private data center and public cloud is another area ripe for improvement.
Whether operators only provide an Internet connection to a cloud service or offer an integrated bundle that includes network services, connectivity needs to be as reliable, responsive and elastic to customers' needs as the cloud itself to make hybrid clouds viable for customers.
Is SDN a solution to hybrid cloud networks?
The reality of networking for hybrid clouds today is that elements in many disparate domains must be stitched together, and they are only able to deliver an approximation of the responsive, elastic platform that hybrid cloud computing customers require. A usable network service combines network elements in several key areas: IP routing; wireless and wireline access, metro and regional aggregation; packet and optical cores and other domains. In most cases, this involves integrating multiple vendors' platforms. But under current implementations, operators are constrained in their ability to support a variety of services with the elasticity and responsiveness required in hybrid cloud computing.
To overcome these complexities, service providers' cloud networks need to evolve and become as responsive and nimble as customers need. Software-defined networking (SDN) may be what operators need to meet some of the challenges inherent in hybrid cloud networking. It accomplishes this using a design similar to the one that's made cloud computing possible -- abstracting software from underlying hardware and unleashing the power of both.
Architecturally, SDN allows network services to be abstractly defined once and "overlaid" broadly onto a population of heterogeneous network elements. Those elements can be in domains as diverse as the ones previously described, such as wireline metro aggregation or optical cores. The control plane -- in which these abstracted services run -- exists between the underlying elements and applications running "above" them in a hierarchical relationship.
By creating this consistently-defined middle tier of network services, operators can simplify and expedite the processes for introducing new offerings. Consequently, providers can overcome some of the key hurdles they face in creating attractive services for hybrid cloud computing customers. They can use SDN to implement a more straightforward method for creating services with latency, capacity, resiliency, security or time-to-activate guarantees tuned to the demands of hybrid cloud computing customers. Returning to the example of Microsoft researchers' experiment with the MapReduce application in a hybrid cloud deployment, an SDN-driven network design could help operators address the latency and capacity issues present in the initial experiment.
Of course, this is the theory of what SDN could bring to customers in the hybrid cloud. Will cloud providers ultimately embrace of SDN to support them? There are early signs that operators indeed find the approach compelling. NTT Communications' SDN-centric Global Enterprise Cloud is already in the market, and trials such as Deutsche Telekom's Terastream project use SDN to enhance its cloud network services. But these are just two examples of innovations in this space. For cloud providers, the key remaining questions will focus not only on what capabilities SDN can deliver in hybrid cloud deployments, but also on which complementary services it should be bundled with and how quickly it can be deployed.
About the author:
Paul Parker-Johnson is the practice lead for cloud computing and virtual infrastructure technologies at ACG Research.