Based on information in this SIGCOMM paper we should be able to provide an indication if the test was limited by the access link or some other link, if it wasn't limited by the send/receive window. We've already seen smoothed RTT values orders of magnitude larger than the minimum RTT when a link is shaped and packets start to queue, so this should be quite possible.
Questions:
- Should we poll TCP_INFO until a retransmit to get slow start information?
- Gather throughput during slow start to compare to overall throughput to see if the situation changed after we stopped polling?
- What values to use for thresholds? Fixed, relative, or try to learn?
Based on information in this SIGCOMM paper we should be able to provide an indication if the test was limited by the access link or some other link, if it wasn't limited by the send/receive window. We've already seen smoothed RTT values orders of magnitude larger than the minimum RTT when a link is shaped and packets start to queue, so this should be quite possible.
Questions: