6533b88bfe1ef96bd12e30e0

RESEARCH PRODUCT

Code for simulating online scheduling of task graphs on hybrid platforms

subject

description

Code used to simulate a hybrid computing platform (consisting of CPUs and GPUs) to test several online algorithms that discover the task graph as it is unveiled and to compare them to the classical HEFT offline scheduler. The related study addresses online scheduling of task graphs on a hybrid platform composed of two types of processors (called CPU and GPU for convenience).This dataset consist of a single .zip artifact file consisting of:/src: source code for the simulator to assess online strategies/simu: source code to launch the simulation and generate plots/artifacts: overview documents with instructions on how to run code and system requirements/Europar: documents relating to the Euro-Par 2018 conference paperThe simulator tests several algorithms to schedule a graph on a platform composed of m CPUs and k GPUs. The related Euro-Par 2018 paper describes the competitive ratio in terms of this m and k.The experimental set-up consists of three instances: - 'Cholesky': realistic DAGs (Directed Acyclic Graphs) corresponding to a linear algebra application: the Cholesky factorization- 'STG' random DAGs generated by the STG set, which is frequently used to compare the performance of scheduling strategies. - 'ad hoc': instances designed to be difficult for this problem and specifically for QA (Quick Allocation - the simplified algorithm reported in the related Euro-Par paper)The experiments run in four steps:1. retrieve the data from STG set2. generate all instances3. run the online strategies4. plot the dataArchived instance files consist of several tabulation-separated columns. Each line describes the result obtained for a given graph file. The first column provides the name of the file, and the other ones provide the makespan obtained by each heuristic.See the related README and overview documents in the artifact for a more detailed description of mehods.Scripts are supplied in R language .R, bash script .sh, c language .c, .h and .cpp formats, instances in .stg files, results in .csv and documentation in text file format, all openly accessible using text edit software. Generated plots and document files are in .tex LaTex format.