|
Contents
There are 4 stages:
- Tuning the toolkit for a concrete architecture
- Preparing the task
- Static tuning of the task for the architecture
- Execution
1. Tuning the system
- Get the file procs.vct defining the performance by processor_test utility
- Get the file link.mtr defining network speed by network_test utility
2. Static tuning of the task for the architecture
- Get the file schedule.sch defining desired processors and desired order of execution.
3. Execution
- Compile the C++ file with system kernel (e.g. by make utility). An output is prog file.
- Execute the program by a standard utility. E.g. mpirun -np 8 prog graph.grf schedule.sch procs.vct link.mtr, where
8 |
number of processors |
prog |
name of file to be executed |
graph.grf |
text representation of algorithm graph |
shedule.sch |
schedule file |
procs.vct |
processor performance file |
link.mtr |
network performance file |
|
|