Leveraging the Strengths of Transactional Memory While Maintaining System Performance for a Multiplayer Gaming Application

摘要

With the technology of multi-core on the rise, many applications have started to support concurrency control on shared resources for raising performance and accuracy such as locks, semaphores, and monitors. A new mechanism called transactional memory (TM) can provide serilizability for using shared resources and ease for programming. However, the TM performance of most studies has been relatively poorer than traditional methods. This is the reason why these applications still use the locks to handle the problem of concurrency control. We propose two mechanisms named Barrier Transactional Memory and Aggregate Physics Update to detect and release signals without conflict and to update all calculation results using parallel multi-threading in a multi-player gaming system. Experimental results show the improvement and the comparison of the proposed mechanisms to traditional locking and single thread update, especially in the area of physics calculations and updating.