Architectural reliability of split-protocol
Bharat S. Rawal, O. Phoemphun, Harold Ramcharan, Lance R. Williams · 2013
HTTP request splitting is a novel concept where the TCP connection and data-transfer phases are dynamically split between servers without the use of a central dispatcher or load balancer. Splitting is entirely transparent to the client and provides security due to the inaccessibility and invisibility of the data servers. This study focuses reliability of Mini Web Server Clusters based on HTTP Request Splitting. Previous work has studied through performance and established a performance improvement range between 6% and 25 % and an improved data transmission time which was 84% better than conventional computing. We believe this higher performance results from inbuilt reliability in the architecture of split protocol. There are plenty of researches, and literatures focusing on component-based reliability. Especially in Software Engineering researches, they assumed that each component has a single Service. Thus, this paper highlights on an inbuilt reliability in a split-protocol due to the dual and interchangeable role of Connection Server (CS) and Data Server (DS). This novel approach introduces a more effective way of offering reliability in cluster computing and in general. To assess the reliability of the system, this paper proposed a simple mathematical model which can capture the reliability of the system. This model is also used to compare the reliability of the split system against dispatcher system. The results show that split system reliability is far better than dispatcher system reliability.