Lightweight Stream Scheduler

From esoterum.org
Revision as of 16:53, 22 April 2011 by Baker (Talk | contribs)

Jump to: navigation, search

Eclipse Projects: Simulation Implementation (ICCD Code)

All Eclipse projects located on agave in /home/mbaker7/workspace/
  • stream_scheduler/
-Debug/
-run_scheduler.sh: Generate schedules based on configuration parameters set in the script
  • StreamIT_compiler: Contains benchmarks used for all scheduler projects, the projects take the cannonical-graph.dot compiler output as their input, the compiler here has been modified to include additional information in the generated dot graphs which are not provided by the original StreamIT compiler

Eclipse Projects: CBE Implementation (Journal Code)

All Eclipse projects located on agave in /home/mbaker7/workspace/
  • scheduler_PPU/
stream_scheduler.cpp: Main source file
compile.sh: Used to compile both PPU and SPU code and embed SPU object into PPU code before sending executable to the CBE
config_exp.h: High level and experimental configurations, applies to both PPU and SPU project
config.h: PPU specific configurations, mostly related to print statements and debugging
  • scheduler_SPU/
config_spu.h: SPU specific configurations, includes print statements, but also determines size of SPU local store buffer which is used for dynamic memory allocation



CBE Infrastructure (cell-user.cc.gatech.edu)

All files located on cell-user.cc.gatech.edu in /home/mbaker/workspace/
  • scheduler_PPU/: Contains executables and scripts for collecting decoding data
-run_scheduler.sh: Experiment script for calling the script to execute ffmpeg using various parameters
-raw_output_data/: Raw scheduler output results should be placed here by the run_scheduler.sh script
-stream_program_source_directories.config: This configuration file is read by the executable, the contents are one benchmark per line including source and input/output file directories and the number of cores to use when running the benchmark, the config file may contain multiple benchmarks to run simultaneously, the total number of SPUs should not exceed 16, or the scheduler will return an error