Feedback Control Quality Adaptation Framework for SEDA System

摘要

The staged event-driven architecture (SEDA) is a new software framework, which is used to construct extreme concurrency Internet services systems. When the system is overloaded maintaining the quality of the requests processing is a difficult problem because both the processing costs and requests arrival rate are highly unpredicted. This paper presents a feedback control quality adaptation framework that adjusts the SEDA system behavior based on the current system status and allows plug-ins of different real-time scheduling policies. An advantage of the framework is its use of feedback control theory as a scientific underpinning. We leverage mathematic methods from the area of control theory to establish dynamic models of SEDA system, which are major challenges and key steps for the design of control theory based system. In this paper, we show how the control-based solution can be utilized to guide load shedding with the target of maintaining requests processing delays.