06-19-2017 11:05 AM - edited 06-19-2017 11:16 AM
I wanted to cover a couple of issues that have come up with how speedtests report link capacity when using GPS Sync, as well as fixed-frame mode without sync.
- Odd asymmetric speedtest results when using the speedtest in the airOS GUI, such as 450 Mbps downlink and 20 Mbps uplink. This is due to an issue when the AP and CPE are in different subnets, which will be fixed in airOS v8.3.1.
- Low speedtest results in general. These are caused by a couple of things:
- TCP window size. Compared to flexible mode, fixed-frame modes will show low throughput if the TCP window size is not large enough due to the bandwidth-delay product. The higher latency seen in fixed-frame mode over flexible mode (though it is still acceptable latency, and by being more consistent provides better performance for VoIP and other applications) will cause a lower speedtest result when using a small TCP window size.
- The number of TCP streams used. If a large-enough TCP window size cannot be used to fill the capacity of the link, the number of TCP streams must be increased.
Note that these TCP speedtest interactions with fixed-frame mode will occur with any fixed-frame product. They are also not an issue with customer's real traffic throughput; when multiple CPEs and applications are transmitting and receiving data they will be able to fill the link capacity.