Many modern massively distributed systems deploy thousands of nodes to cooperate on a computation task. The underlying network is becoming the bottleneck in these systems and network congestion happens. System designers have to deploy protocols with congesiton control algorithms to protect the systems from congestion collapse. Most of the existing congestion control algorithms are loss-based algorithms. The commonly used protocols are TCP for reliable communication and TFRC for unreliable communication. Both TCP and TFRC use packet loss as signal to detect congestion.In this paper, we study the packet loss process in sub-RTT timescale and its impact on the loss-based congestion control algorithms. Our study suggest that the packet loss in sub-RTT timescale is very bursty. This burstiness leads to two effects. First, the sub-RTT burstiness in loss process leads to complicated interactions between …

Abstract: Contents1 Executive summary 32 Motivation 42.1 Demand for ultrascale networking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42.2 TCP/IP paradigm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42.3 Protocol scalability problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 Pilot project: performance and stability 53.1 Network model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ....