The Research Of Multi-level Real-time Network Stream Scheduling On Multicore Network Processor

As the continuous development of Internet, video and multimedia services based on network have made great progress, the video service will become the core of Internet applications. Video service system demands a lot of computing resource to provide high concurrency of high-quality real-time network data streams, general purpose processors are unable to meet the requirement of performance. Because multicore network processor not only has excellent computing capacity but also integrates lots of coprocessors to accelerate IP packets processing, multicore network processor has been the preferred alternative for high performance application platforms. However, real-time video streams are real-time and fluctuant, so that how to take full advantage of multicore network processor to guarantee the smooth transmission of video streams, and reduce the impact on network of fluctuation of streams transmission will be the key issue which will influence the direction of future Internet development.In order to resolve the problem described above, this dissertation presents real-time network stream scheduling platform based on multicore network processor, and researches into depth of three levels which are task scheduling among cores, multicore topology and processors resource cooperation. The main work and contributions of the dissertation are as follows:(1) This dissertation presents a multi-level real-time network stream scheduling framework on multicore processor, which integrates task scheduling among cores, multicore topology management and processors resource cooperation strategy from three levels, so that the framework can optimizes the performance of real-time network stream scheduling in different granularity and take full advantage of capacity of processing and parallel characteristic of multicore processor;(2) This dissertation develops a real-time network streams with variable period scheduling model by real-time and non-realtime processing mixed and collaborating, and the model transforms streams processing with variable period into hybrid task scheduling on multicore processor. The model enhances the throughput and concurrent of the system on the condition that the streams can be scheduled with real-time characteristic;(3) We propose EDF-segment algorithm to solve hybrid task scheduling problem with variable period. EDF-segment algorithm regulates real-time tasks on multicore processor by online dynamic programming strategy, which eliminates time slices caused by real-time task scheduling. The algorithm increases efficiency hybrid task scheduling on multicore and resolves the multi-objective optimization problem of multicore processor utilization and real-time streams scheduling;(4) A fluctuant model for analyzing real-time network stream sending is brought out, by which we develop real-time network stream scheduling platform to achieve smooth transmission for concurrent streams;(5) We also improve real-time resource management and load migration strategy on multi-processor to solve the problem of the waste of computing resource or lack of computing resource caused by network stream jitter, and the strategy introduces a prediction mechanism for load migration, and settles processor resource unbalance and tasks processing delay caused by stream jitter, so that the capacity of real-time network stream scheduling framework for real-time stream processing is exploited thoroughly.