The basic problem limiting TCP performance arises from how the TCP congestion avoidance algorithm interacts with networks having large BDP. Congestion avoidance increases the sender's TCP window by only a single packet for each successful round-trip acknowledgment.
When the TCP window is small, increasing it by a single packet is reasonable, but if a window is very large (for hundreds of packets), then each additional round-trip acknowledgment adds a small increase to the sender's TCP window. In this case, it takes an extraordinarily large number of round trips to rebuild the TCP window in response to a single packet loss, and this leads to slow TCP behavior.
Data shows that when the bandwidth is above a specific level (4M), it has no effect on the webpage loading. On the other hand, reduced RTT directly benefits page loading.
During the TCP congestion avoidance phase, lower RTT decreases the BDP threshold, which means higher throughput for the same bandwidth and RWIN values. The throughput is then limited by the segment with the highest RTT.
AA TCPO reduces RTT, as retransmissions because of loss in the access network are handled by the AA instead of the content server. AA allocates enough buffers in the downlink direction according to the BDP.
The main benefit of reducing RTT is a faster recovery from a packet loss event.