A TCP Incast Avoidance Method Based on Retransmission Requests from a Client

When a standard TCP implementation using minimum retransmission timeout (RTOmin) of 200 ms is used in distributed file systems in data centers, a well-known throughput degradation called TCP Incast occurs, because 200 ms is too large as RTOmin in data centers. In order to avoid TCP Incast, a TCP implementation using a much smaller RTOmin attained by a fine-grained kernel timer has been proposed. However, throughput still decreases in the implementation when the number of servers in distributed file systems is large. In this paper, we propose an improvement method using special acknowledgement packets which enable a client to request servers to retransmit packets in a better timing. Numerical examples obtained by extensive simulation runs show the effectiveness of the improved method.