Fast Intercept of Passing Streams for High Performance Appliances
in Application Service Networking
Internet is increasingly being "active". Documents are being dynamically processed before they are served. The location of processing is also dynamic. The work investigates two aspects: (a) how documents can be processed within an overall service model/scenario in any location between the origin server and the user-agent; (b) what type of network software layer in the intercepting machine can expedite intermediate information processing. Random access into processed data is believed to be an important performance criterion in any computation. Envisioning a generalized framework for supporting a wide range of possible content services, the thesis suggests a novel content scoping and indexing based random access mechanism into a passing stream for intercepting filter like appliances on this framework. It also presents an application programming interface for efficient stream editing. The work also presents a user space implementation of the proposed intercepting machine and a performance study of the scheme on this implementation. Even without any kernel level support, the implementation showed about 500-800% speedup over today’s content servicing technique in normal conditions. The result suggests such random access can significantly speed up future intercepting applications of Internet. |
![]() |