Add docker image with CPU only dependencies

[johnugeorge]

In this PR

1. Versions of packages are frozen in requirements file for reproducibility
2. CPU version of Dockerfile is added with no GPU dependencies

Docker file installation steps match the GitHub action DLIO installation

[johnugeorge]

@jovonho

[lhovon]

Looks good to me!

[zhenghh04]

@johnugeorge Are you able to check whether the dockerfile works in MacOSX?

When I was running with bare metal version, there was some issue with iostat profiling in MacOSX system. Let us make sure this works before adding that to the dockerfile.

[johnugeorge]

@johnugeorge Are you able to check whether the dockerfile works in MacOSX?

When I was running with bare metal version, there was some issue with iostat profiling in MacOSX system. Let us make sure this works before adding that to the dockerfile.

@zhenghh04 Sorry, I didn't catch your comment on MacOSX. IOStat is installed inside the debian docker container right? I assume, Github actions VM image already has IOStat installed in it for profiling.

dlio_benchmark/.github/workflows/python-package-conda.yml

Line 7 in 19e2816

runs-on: ubuntu-latest

I was using MacOSX to build the image. I tried

dlio_benchmark/.github/workflows/python-package-conda.yml

Lines 46 to 47 in 19e2816

	mpirun -n 2 python ./src/dlio_benchmark.py workload=unet3d ++workload.framework=tensorflow ++workload.data_reader.data_loader=tensorflow ++workload.train.computation_time=0.05 ++workload.evaluation.eval_time=0.01 ++workload.train.epochs=2 ++workload.workflow.train=False ++workload.workflow.generate_data=True ++workload.dataset.num_files_train=16 ++workload.dataset.num_files_eval=4
	mpirun -n 2 python ./src/dlio_benchmark.py workload=unet3d ++workload.framework=tensorflow ++workload.data_reader.data_loader=tensorflow ++workload.train.computation_time=0.05 ++workload.evaluation.eval_time=0.01 ++workload.train.epochs=2 ++workload.workflow.train=True ++workload.workflow.generate_data=False ++workload.dataset.num_files_train=16 ++workload.dataset.num_files_eval=4

inside docker and it works. Anything else to be verified?