While WAN-compression solutions have been around for years, new compression advances have resulted in previously unheard of gains in bandwidth savings. Delta compression, commonly referred to as segment caching or byte caching, leverages pattern-matching techniques and large persistent dictionaries to dramatically reduce the amount of data sent across the WAN.

Delta-compression systems are symmetric, which means they require components, either software or hardware, on both ends of the network. In almost all cases, the server-side component is a dedicated appliance and the client-side component is either a software module installed on the PC or an appliance deployed in the data path.

Software client-based systems have the advantage of requiring hardware on only one side of the link, making the approach suited for deployments in which there are only a few users per location. However, this flexibility comes at a cost. Client-based compression systems are limited in that they operate only on the data sent to that particular client. A file downloaded by one client therefore provides no benefit to other users. Furthermore, client-based compression systems require an additional download during initial application access. This download degrades first-access performance.

While appliance-to-appliance delta compression requires hardware at both ends of the network, it offers significant performance advantages over client-based deployments. First, appliance-to-appliance delta compression allows cross-user benefit. When one user downloads a file, the transferred bytes can be used to compress the same file when it is requested by a second user. Additionally, symmetric appliance deployments have no first-transfer penalty because no client code has to be installed. Finally, symmetric appliance deployments provide benefits not found in client-based systems, such as QoS capabilities.

In addition to hardware and software deployment techniques, delta-based compression can be achieved at different network layers. Some systems operate at lower layers of the OSI model, while others operate at higher layers. The layer of operation has a significant impact on compression effectiveness.

IP or TCP (Layer 3 or Layer 4) compression systems buffer packets that are then compressed one at a time or as a group and sent to the decompressor. The primary problem with Layer 4 compression is that it compresses packets individually or mixes different data types together. Both of these options can negatively affect the final compression ratio.

One alternative to packet-level compression is application layer-based (Layer 7) compression. This approach involves buffering the server's response and then computing a delta between the current response and a reference response. This reference response can be thought of as a typical response for the current request.

This approach improves compression ratios by extending the scope of the delta operation from the packet to the entire application response, but limits the benefits to one protocol. Furthermore, application-layer compression only performs a delta between a single reference response and the current one. As a result, the amount of data that can be drawn from is limited, which dramatically impacts overall compression ratios.

The ideal compression solution operates at the session layer (Layer 5). This allows it to apply compression across a completely homogeneous data set while addressing all application types. Furthermore, session layer operation eliminates packet boundary limitations. This makes it easier to find long matches in the datastream.

Overall, the most critical consideration when comparing delta-compression systems is throughput. While achieving a high compression ratio is important, maximum throughput is vital to improving application performance. Delta-compression systems often achieve compression ratios of 95 percent or higher. To fully translate this bandwidth reduction into performance, the system needs to operate at 20 times the WAN speed.

In conclusion, the performance gains from a given compression technology can be assessed by considering the technology's expected compression ratio, the devices' peak compression throughput and the network bandwidth. Too low a compression ratio and the network will remain saturated and performance gains will be minimal. Similarly, too low a compression speed, and the compression device itself becomes the bottleneck.

Saxon Amdahl is a technical marketing manager at F5 Networks.