The notion that WAN optimization is best attacked at a single layer of the Open Systems Interconnection model – Layer 5, the session layer – is intriguing. However, reality suggests otherwise. WAN performance is affected by many factors, each of which has unique characteristics that must be addressed at different OSI layers to have the greatest possible impact. Approaching the problem at too low or too high a level produces sub-optimal results.
The other side: Layer 5 is the answer
The forum – what do you think?
For WAN optimization, problems should be resolved at the lowest possible layer to achieve the broadest applicability across the greatest number of applications with the least amount of complexity. Consider the two most pressing issues affecting WAN performance today: bandwidth and latency.
Solving the bandwidth problem requires reducing the amount of traffic crossing WAN links to free up additional capacity. The key to reducing WAN traffic is pattern matching – preventing redundant data from consuming valuable bandwidth by sending it once, then replacing repetitious transmissions with a tag or label.
For pattern-matching to be effective, WAN optimization products must look across multiple sessions and users to identify and remove the greatest number of repeated data patterns. While pattern-matching at Layer 5 will identify redundant patterns within a single session, it will miss repeated patterns across different sessions.
For maximum effectiveness, pattern-matching is best performed at Layer 3, the network layer, where algorithms can be applied that recognize patterns of any size, across multiple sessions and applications – including User Datagram Protocol (UDP )-based applications such as VoIP. By optimizing at a lower, protocol-independent layer, pattern matching has the greatest possible impact across the widest range of WAN traffic while being completely transparent to existing network devices and servers.
When it comes to latency, there are two issues to address, both of which require different approaches at different OSI layers. First, latency affects the performance of TCP-based applications at Layer 4, the transport layer. Because TCP restricts the amount of data sent across the WAN, traditional approaches aren’t always sufficient to improve performance. The ideal solution must seamlessly replace TCP at Layer 4 with a more efficient protocol designed specifically to deal with latency and accelerate TCP-based applications.
WAN latency impacts “chatty” applications, which require hundreds or even thousands of time-consuming round trips between the client and server to complete a transmission. These applications require protocol-specific acceleration to minimize chattiness and speed up performance across the WAN. This is best addressed at the top of the OSI stack – Layer 7, the application layer.
Clearly, no single OSI layer offers a magic tonic for WAN optimization. An approach that provides broad coverage with the least amount of overlap is a much better strategy.
