Hybrid cloud is getting a lot of attention, but working in a hybrid cloud environment is still a fairly rare thing,...
according to Nemertes Research Group Inc. Most companies now have a hybrid service delivery environment, but that just means technology solutions are being provided by some combination of internal and external resources. A hybrid cloud requires two elements: a cloud on the inside, in your data center space, and a cloud on the outside, into which you can migrate workloads as needed. Such an environment creates hybrid cloud optimization challenges for networks.
Hybrid cloud optimization is a goal for many IT shops … [but] such an environment … brings with it new kinds of network performance challenges.
Here are a few:
Python swallowing a pig -- One challenge, of course, is the whole issue of moving a virtual machine (VM) from inside the data center to some infrastructure as a service resource pool. A VM is typically very large. In any kind of production situation, it could easily interfere with other kinds of traffic as it pushes through the pipe, increasing the latencies and even packet losses on other data flows as their packets get crowded out or stuck behind the VM in motion.
Traffic shaping appliances or even router quality-of-service settings might allow IT to prioritize the massive transfer so that it ranks lower than other functions. That may ensure those other functions don't starve, but it could also interfere with the VM migration completing in a timely fashion. Path optimization -- provided by physical appliances, virtual appliances, a mix or a cloud bridge system -- could compress the VM dramatically, especially if it is built on a well-defined "golden image" used by other VM images.
Opening the spigot -- Another hazard in the hybrid cloud is the shifting of service components, whether in the form of a VM or in the form of an application running on an app server or database, from internal resources to external, or vice versa. As the component moves, the stream of transactions it is processing will have to move as well to follow it. This could result in significantly changed data volumes across the wide area connection between the resources, be it Internet or dedicated link. Additional latency and packet loss can also occur if that shift not only pulls new streams of data across that link but inserts round trips across it as well.
Wide area network optimizers armed with protocol spoofing capabilities and data compression features might be able to mitigate packet loss and latency by circumventing some round trips and decreasing data volumes. The best optimization, of course, would be to foresee the problem. Careful performance monitoring and component mapping would let IT model the changes in traffic flows that would follow component motion, thus avoiding rather than merely mitigating the problems.
Autoscaling -- Another hybrid-cloud-specific consideration is the sudden shift in traffic flows that can follow automatic scaling of services. One of the most important features of a robust cloud environment is that it can, in response to demand, expand service capacity on the fly. If, for example, demand for a website doubles, the cloud can spin up new front-end and application servers to meet the need. If some of the components being scaled up are on the outside and others are on the inside, then the flows across the link could suddenly increase -- detrimentally to other traffic. Again, prioritization can be the key technology here to balance out the demand.
Hybrid cloud optimization is a goal for many IT shops, and the hybrid cloud's promise for flexibility, agility and robustness is very exciting. Such an environment, however, brings with it some new kinds of network performance challenges, and IT departments have to be prepared to confront them as they make strategies and lay plans for a hybrid cloud environment.