Background: the benchmark framework is too messy to facilitate increasing benchmarking complexity.
Basic requirements for the bencher:
- Argument parsing
- Launch the VM
- Execute the workload
Aim: make the bencher more modular. There are mainly two types of modules:
- Modules for different variant of kernel
- Modules for different kinds of workload
Put different kernel and workload supporting code into separated modules. Then abstract away the common workflow among them.
Argument parsing
Currently all argument parsing is done inside cli.py,
additional arguments for each workload has to be specified in the configure() function.
VM lanunching
VM launching needs also to facilitate subtleties acrossing kernels:
- Different commandline for kernel boot;
- Different level of support for
virtiofs. This is the most fundamental requirements to properly run a benchmark. Because all required resources are stored on the host and mounted as virtiofs; - Different kernel module or drivers to be inserted