Context Aware Concurrent Execution Framework forWeb Browsers

摘要

Computing hungry multimedia web applications need to efficiently utilize all the resources of a device. HTML5 web workers is a non-sharing concurrency platform that enables multimedia web application to utilize the available multi-core hardware. HTML5 web workers are implemented by major browser vendors to facilitate concurrent execution in web clients and enhance the quality of ambitious web applications. The concurrent execution in web workers allows parallel processing using available cores at the expense of communication overhead and extra computation. The benefits of concurrent execution can be maximized by balancing load across workers/CPU cores. This work presents load-balancing algorithms between web workers using parameters such as scheduler throughput, computation priority and game entities locality. An award-winning web-based multimedia game (raptjs.com) is used to evaluate the performance of load balancing algorithms. The preliminary results indicated that the performance of game improved with the proposed load-balancing across web workers. The load balancing algorithms were developed on top of DOHA [1], an open source JavaScript execution layer for multimedia applications. The load between web workers is transferred between web workers via serialized objects. Effects of load transfer between on the overall application performance was measured by analysing jitter and number of frames per second. Load balancing algorithms and load transfer mechanism can be used by developers to improve the application design, and giving end users better perceived quality of experience for demanding multimedia web based applications.