Optimizing Bandwidth for Fun and Profit

Masters of

Bandwidth optimization is now a mainstream technology with more than 61% of companies using the technology to improve application performance and/or reduce bandwidth requirements.

Still, some IT architects ask, “why can’t I just use QoS technologies like low latency queuing or class-based weighted fair queuing to meet application performance requirements?”

If performance, and not bandwidth, is the only problem, and if the performance problems relate only to real-time communications versus other traffic, it may be possible to address those problems using existing class-of-service/quality-of-service features in core network routers. By putting communications traffic in a "real-time" priority class, for delivery before other packets, many organizations can address problems with intermittent packet loss and jitter in their voice and video traffic.

However, QoS systems are not very granular - they typically only allow four classes of service, for example. If the organization needs finer grain control of their traffic, then QoS alone will not solve the problem. Also, there are kinds of latency that QoS can't help with: QoS does nothing to ameliorate the latency introduced by distance, and cannot reduce latency associated with delivery of large packets. Optimization systems, on the other hand, can provide additional granularity, and can sometimes mitigate these other forms of latency as well.

Another great benefit, apart from the problems that optimization can solve that adding bandwidth cannot, is cost reduction. An organization can usually cost-justify deploying WAN optimization just by looking at the ability to avoid or delay bandwidth upgrades. Consider a typical organization with a few T1-level WAN links. Faced with a need for more WAN capacity, they planned to bump up from T1 to T3 rates, at a quadrupling of costs. By bringing in WAN optimization instead, even including all the costs of testing, installing, and maintaining the gear, they saved about $130,000 in the following year! (These savings are especially pronounced in underserved markets where bandwidth is still scarce and very expensive.)

Optimization uses various techniques to reduce traffic volume and improve application performance: content caching and data compression reduce volume, while protocol accelerations, prioritization, and traffic shaping control and improve application performance over the network. Accelerators, either by helping the network avoid packet drops or by tinkering with the TCP flows directly and adjusting drop-back/ramp-up behaviors, mitigate this kind of performance problem.

Modification of flow controls on TCP streams is one of the techniques used to shape traffic. Traffic shaping aims to ensure that an organization has some control over how applications consume available bandwidth. Control can be positive, guaranteeing that certain applications, devices, or users get bandwidth, or negative, limiting the bandwidth that specific users, devices, or applications get. Other techniques include queuing (parking packets in buffers, or queues, to await delivery after higher priority packets), connection-closing (sending a “close” packet to one or both ends of a stream, to shut it down), and selective packet dropping (creating artificial packet loss, to stop ramp-up or force step-down on a TCP stream's speed).

Optimization Architectures

Those deploying WAN optimization must choose between symmetric and asymmetric architectural models.

Symmetric optimization requires an optimizer at both ends of a connection to function. The end points can be dedicated appliances, embedded functions in a multi-function device, or even software running on servers or on a user's computer. Because symmetric optimizers control both ends of a traffic stream, they can do more significant compressions and accelerations.

However, putting something at both ends of connections, especially when that something is dedicated hardware, adds cost and complexity that grows as the network grows, increasing the management burden of IT. Soft clients may or may not cost extra on a per-seat basis, but because they usually run on a PC, they increase complexity and support burdens in the form of helpdesk calls.

Asymmetric optimizers only control one end of a connection. Optimizations that don't require special clients on the other end include some protocol accelerations, traffic shaping, and compression of Web content, since Web browsers have the ability to decompress content which has been compressed, using a few standard techniques. Many asymmetric optimizers incorporate features aimed at offloading work from Web servers, and for load balancing among them. Some include security functionality as well, such as application firewalls for Web apps.

Because they are deployed only in the data center end of the network, asymmetric optimizers don't have the same capital or management overhead that come with broad deployments of symmetric devices and they can serve arbitrary end users, not just those with company computers. However, they cannot compress traffic as much or perform all the accelerations possible with symmetric devices.

The bottom line: if you aren’t evaluating WAN optimization, you should. Our research indicates that those who deploy it save money, improve application performance, and in 95% of all cases, rate their deployments as “extremely” or “very” successful.

Join the Network World communities on Facebook and LinkedIn to comment on topics that are top of mind.

Copyright © 2011 IDG Communications, Inc.