Running programs on Hydra
The Cluster Machine has 16 nodes (0 to 15) plus one master node (-1). User programs
MUST NOT be run on the master node but on compute nodes instead.
Before you run your programs, please be sure to check the cluster
status by using beostatus –c command (see Lab 1 for more details) to
select a non-busy node (nodes with most CPUs with 0% usage). Running
programs on a busy node may result in incorrect performance benchmark
and you will not receive credit for this lab.
bpsh command
- To run programs on hydra, use the
bpsh command.
- You can run C, C++, openMP, cuda and Fortran programs on Hydra through bpsh command.
- bpsh will explicitly
define on which node the program will run.
-
See at command in order to schedule jobs on hydra.
Syntax:
bpsh n <path> exec_code arg1 arg2
or
bpsh n ./exec_code arg1 arg2
where,
n is the node number (n =
0, 1, 2, 3 … 15)
arg1, arg2, arg3…: program
arguments
Example:
bpsh 5 /home/john/csc5551/cge 8 1
OR
bpsh 5 ./cge 8 1
This command executes the cge program on node 5, where arg1 is 8 and
arg2 is 1.
Running C/C++/openMP and Fortran programs on Hydra
You can run C/C++/openMP and Fortran programs only on nodes 0, 1, 2, 3, 5, 6, 7, 8, 9, 10, 11, 12, 13 and 15.
Do not use nodes 04 and 14 because they are not working.
Do not use node -1 because it is the master node.
Running Cuda programs on Hydra
Cuda programs can only be executed on nodes 12 and 13 because these nodes are connected to the GPU.
Cuda programs need GPU in order to execute.