Nonpreemptive run-time scheduling issues on a multitasked, multiprogrammed multiprocessor with dependencies, bidimensional tasks, folding and dynamic graphs
Increases in high speed hardware have mandated studies in software techniques to exploit the parallel capabilities. This thesis examines the effects a run-time scheduler has on a multiprocessor. The model consists of directed, acyclic graphs, generated from serial FORTRAN benchmark programs by the parallel compiler Parafrase. A multitasked, multiprogrammed environment is created. Dependencies are generated by the compiler. Tasks are bidimensional, i.e., they may specify both time and processor requests. Processor requests may be folded into execution time by the scheduler. The graphs may arrive at arbitrary time intervals. The general case is NP-hard, thus, a variety of heuristics are examined by a simulator. Multiprogramming demonstrates a greater need for a run-time scheduler than does monoprogramming for a variety of reasons, e.g., greater stress on the processors, a larger number of independent control paths, more variety in the task parameters, etc. The dynamic critical path series of algorithms perform well. Dynamic critical volume did not add much. Unfortunately, dynamic critical path maximizes turnaround time as well as throughput. Two schedulers are presented which balance throughput and turnaround time. The first requires classification of jobs by type; the second requires selection of a ratio value which is dependent upon system parameters. 45 refs., 19 figs., 20 tabs.
- Research Organization:
- Illinois Univ., Urbana (United States). Center for Supercomputing Research and Development
- Sponsoring Organization:
- USDOE
- DOE Contract Number:
- FG02-85ER25001
- OSTI ID:
- 5546139
- Report Number(s):
- DOE/ER/25001-45; UILU-ENG-87-8002; CSRD-656; ON: DE88003598
- Resource Relation:
- Other Information: Thesis. Portions of this document are illegible in microfiche products
- Country of Publication:
- United States
- Language:
- English
Similar Records
Data Locality Enhancement of Dynamic Simulations for Exascale Computing (Final Report)
Generalized multiprocessor scheduling for directed acylic graphs
Related Subjects
ARRAY PROCESSORS
PARALLEL PROCESSING
TASK SCHEDULING
ALGORITHMS
BENCHMARKS
COMPUTER GRAPHICS
COMPUTERIZED SIMULATION
FORTRAN
OPERATION
PERFORMANCE TESTING
DATA PROCESSING
MATHEMATICAL LOGIC
PROCESSING
PROGRAMMING
PROGRAMMING LANGUAGES
SIMULATION
TESTING
990210* - Supercomputers- (1987-1989)