## implemented
- 1 to n engines per group
- 1 to n threads running on one specific core / dsa engine
- copy inside and across NUMA borders
- cross-copy: 2 engines copying from their numa domain to the domain of the other
- all with "packet sizes" of 1KiB, 2KiB, 4KiB, 8KiB, ..., 1GiB
- all with both CPU and DSA for comparison
## missing
- batch vs single submissions
- effect of fence/drain