Live Migration versus Pre-Positioning in the Cloud

The secret to live migration isn’t just a fat, fast pipe – it’s a dynamic infrastructure

Very early on in the cloud computing hype cycle we posited about different use cases for the “cloud”. One that remains intriguing and increasingly possible thanks to a better understanding of the challenges associated with the process is cloud bursting. The first time I wrote about cloud bursting and detailed the high-level process the inevitable question that remained was, “Well, sure, but how did the application get into the cloud in the first place?”

Back then there was no good answer because no one had really figured it out yet.

Since that time, however, there have grown up many niche solutions that provide just that functionality in addition to the ability to achieve such a “migration” using virtualization technologies. You just choose a cloud and click a button and voila!

Yeah. Right. It may look that easy, but under the covers there’s a lot more details required than might at first meet the eye. Especially when we’re talking about live migration.

LIVE MIGRATION versus PRE-POSITIONING

Many architectural-based cloud bursting solutions require pre-positioning of the application. In other words, the application must have been transferred into the cloud before it was needed to fulfill additional capacity demands on applications experiencing suddenly high volume. It assumed, in a way, that operators were prescient and budgets were infinite. While it’s true you only pay when an image is active in the cloud, there can be storage costs associated with pre-positioning as well as the inevitable wait time between seeing the need and filling the need for additional capacity. That’s because launching an instance in a cloud computing environment is never immediate. It takes time, sometimes as long as ten minutes or more. So either your operators must be able to see ten minutes into the future or it’s possible that the challenge for which you’re implementing a cloud bursting strategy (handle overflow) won’t be addressed by such a challenge.

Enter live migration. Live migration of applications attempts to remove the issues inherent with pre-positioning (or no positioning at all) by migrating on-demand to a cloud computing environment and maintaining at the same time availability of the application. What that means is the architecture must be capable of:

Transferring a very large virtual image across a constrained WAN connection in a relatively short period of time
Launch the cloud-hosted application
Recognize the availability of the cloud-hosted application and somehow direct users to it
When demand decreases you must siphon users off (quiesce) the cloud-hosted application instance
When no more users are connected to the cloud-hosted application, take it down

Reading between the lines you should see a common theme: collaboration. The ability to recognize and act on what are essentially “events” occurring in the process require awareness of the process and a level of collaboration traditionally not found in infrastructure solutions.

CLOUD is an EXERCISE in INFRASTRUCTURE INTEGRATION

Sound familiar? It should. Live migration, and even the ability to leverage pre-positioned content in a cloud computing environment, is at its core an exercise in infrastructure integration. There must be collaboration and sharing of context, automation as well as orchestration of processes to realize the benefits of applications deployed in “the cloud.” Global application delivery services must be able to monitor and infer the health at the site level, and in turn local application delivery services must monitor and infer the health and capacity of the application if cloud bursting is to successfully support the resiliency and performance requirements of application stakeholders, i.e. the business.

The relationship between capacity, location, and performance of applications is well-known. The problem is pulling all the disparate variables together from the client, application, and network components which individually hold some of the necessary information – but not all. These variables comprise context, and it requires collaboration across all three “tiers” of an application interaction to determine on-demand where any given request should be directed in order to meet service level expectations. That sharing, that collaboration, requires integration of the infrastructure components responsible for directing, routing, and delivering application data between clients and servers, especially when they may be located in physically diverse locations.

As customers begin to really explore how to integrate and leverage cloud computing resources and services with their existing architectures, it will become more and more apparent that at the heart of cloud computing is a collaborative and much more dynamic data center architecture. That without the ability not just to automate and orchestrate, but integrate and collaborate infrastructure across highly diverse environments, cloud computing – aside from SaaS - will not achieve the successes it is predicted.