The linked article doesn't provide enough information to understand how they're using Acks to probe bottleneck bandwidth, but they've prior work on this. If it's similar to Van Jacobson's pathchar, I would have thought there might be "interesting" interactions with NICs that do TCP TSO and GRO (using the NIC to do TCP segmentation at the sender and packet aggregation at the receiver), as these will mess with timing and, in particular, how many acks get generated. Still, the authors are very well known and respected in the Networking community, so I expect they will have some way to handle this.