National Laboratory and UC

Some TCP congestion control algorithms [1, 34] can provide impressive performance with the help of Explicit Congestion Notification (ECN) or improved timestamping. However, configuring switch queues to mark ECN appropriately or separating low latency protocols from aggressive legacy traffic is not always feasible. And the inertia and variety of networks makes modifying both senders and receivers, altering drivers, or adding new TCP options a daunting challenge. We propose TCP Inigo, a delay-based congestion control variant of TCP that has low barriers to deploy and can coexist with loss-based TCPs. Inigo adds two independent types of congestion control: a Round-Trip-Time (RTT) based congestion ratio on the sender, and a Relative Forward Delay (RFD) based congestion ratio used by the receiver to either mark ECN or manage the receive window. We compare and contrast these uses of RTT and RFD with ECN, demonstrating more than 2× smaller RTTs than non-ECN TCPs and superior utilization in easily reproduced Mininet experiments.