Application Development Framework for Performance-Oriented Multi-Threaded Application Development

Abstract

In this research project, I propose a new approach for developing performance-oriented platform-independent infrastructure layer for the development of concurrent C++ applications with the intention of allowing the programmers to develop high-quality low-latency concurrent applications easily and free of concurrency issues. The project focuses on developing a framework in which traditional master-slave threading model has been extended together with asynchronous communication to support non-blocking synchronization between working threads. The framework consists of several sub modules, which can be used to write multi-threaded performance-oriented applications which will utilize the research outcomes. With the main intention of improving the performance, I have derived some data structures and adaptive asynchronous inter-thread data communication mechanisms which will guarantee to outperform the traditional memory management and inter-thread data communication mechanisms. I have designed a deferred update based software transactional memory module in C++ which is the key component in my research and I have shown how lock-free reference counting based garbage collection can be applied to the memory management of shared objects. I have integrated the transactional memory module to the framework which makes the non-blocking shared memory access possible between threads. As a mean of validating the project success, I have also presented a set of simple preliminary performance tests which will show the performance gain achievable by the designed communication mechanisms and data structures. Further, as one of the main objectives of this project is to use the framework in Automated Trading Systems in order to have a boost in performance, I have integrated the framework to the architecture of an actual Automated Trading System and obtained positive results for the achievable performance gain. The performance benchmarks that will be presented in this report will provide a quantitative proof for the successful outcome of the project.