Middleware for transparent TCP connection migration : masking faulty TCP-based services

Abstract

Mission critical TCP-based services create a demand for robust and fault tolerant TCP communication. Sense Intellifield monitors drill operations on rig sites offshore. Critical TCP-based services need to be available 24 hours, 7 days a week, and the service providers need to tolerate server failure. How to make TCP robust and fault tolerant without modifying existing infrastructure like existing client/server applications, services, TCP stacks, kernels, or operating systems is the motivation of this thesis. We present a new middleware approach, first of its kind, to allow TCP-based services to survive server failure by migrating TCP connections from failed servers to replicated surviving servers. The approach is based on a proxy technique, which requires modifications to existing infrastructure. Our unique middleware approach is simple, practical, and can be built into existing infrastructure without modifying it. A middleware approach has never been used to implement the proxy based technique. Experiments for validation of functionality and measurement of performance of the middleware prototype are conducted. The results show that our technique adds significant robustness and fault tolerance to TCP, without modifying existing infrastructure. One of the consequences of using a middleware to make TCP communication robust and fault tolerant is added latency. Another consequence is that TCP communication can survive server failure, and mask it. Companies providing robust and fault tolerant TCP, is no longer dependant of third party hardware and/or software. By implementing our solution, they can gain economical advantages. A main focus of this report is to present a prototype that demonstrates our technique and middleware approach. We present relevant background theory which has lead to the design architecture of a middleware approach to make TCP communication fault tolerant. Finally we conduct experiments to uncover the feasibility and performance of the prototype, followed by a discussion and conclusion.