Motivation Application semantics do not directly map to priority based scheduling

1. Motivation Application semantics do not directly map to priority based scheduling Application based state information should be utilized so that behaviour can adapt to the situational context Video processing pipelines form the driving application The KURT Linux Group Scheduling framework allows application specific scheduling decisions The new programming model is a general solution to expressing a wide range of scheduling semantics using state and control variables A Programming Model for Precise Computation Control of Multiple Video PipelinesJayanth Venkataraman, Dr.John Gauch, Dr.Douglas Niehaus (niehaus@ittc.ku.edu) Overview Video Pipeline Structures • A Video pipeline structure looks as follows • Video processing pipelines perform video based computations at each stage of the pipeline • KUIM Video processing pipelines were used as the driving application • Certain applications require the balanced progress of multiple streams of video • Timed playback of video in order to exhibit precise reproducibility of recorded video is another KUIM based experiment that uses the programming model Programming Model Design Video Pipeline • Scheduling decision functions (SDF) that are part of the group scheduling framework form the core component • SDF's are designed according to application level semantics and depend on the state level information • State variables represent the state level information of each stage of the pipeline • Control Variables, derived from the pipeline behavior by the SDF, can exhibit control over the pipeline algorithm • The interaction of data between the user space program and the kernel based SDF can be done by system calls or by the use of shared memory Stage 1 Stage 2 Stage 3 Stage 4 State and Control Variables Linux Scheduler Group Scheduling Framework Scheduling Decision Function Members representing each stage of the pipeline