Acceleration can be implemented in three places: on the server, at the desktop or in the network. With accelerators on the servers, each server is ignorant of the WAN capacity available in real time that can be used to accelerate application delivery to users. Because the WAN is shared with other servers and applications, server-based acceleration can be suboptimal at best.

In addition, the load on servers from doing application acceleration simultaneously with computation makes this approach less attractive. Acceleration at the desktop has some of the same weaknesses as server-based acceleration. Furthermore, application-acceleration techniques benefit from the fact that multiple users in a location typically access the same data. These benefits are lost when each desktop tries to accelerate its own data.

Face-off: The case against application acceleration

By deploying acceleration at the network level, organizations can optimize performance equitably by distributing intelligence across the entire infrastructure. But an application-acceleration product that is merely bolted onto the network rather than built into it could undermine the network's health and efficiency and the business processes that run on it. When looking to enhance performance, always heed the old adage: First, do no harm.

To avoid that harm, the acceleration product should be logically integrated into the network infrastructure transparently - in such a way that critical services and functions aren't disrupted or disabled, and IT intervention is kept to a minimum. The biggest problem with nontransparent application-acceleration products is that they force a company to overlay a second network on top of the existing network. This not only adds unnecessary complexity but also can interfere with carefully crafted routing designs and the ability to discover and monitor network devices.

Because an overlay acceleration network focuses on controlling the direction of application packet flows, it could cripple the existing network's power to choose the fastest paths for all the traffic flowing between nodes.

An accelerator solution that's an integral, transparent part of the network infrastructure will preserve critical header information and not cause problems for existing services. Autodiscovery of all the accelerators makes it possible to negotiate and apply an optimization policy to the application flow without disrupting the underlying routing network. Intermediate accelerators along the path can go into pass-through mode, allowing the accelerators closest to the communicating nodes to facilitate the flow optimization. You can't do any of that if an overlay network is in the way.