Spurious TCP Timeouts in 802.11 Networks

Abstract

Abstract-In this paper, we investigate spurious TCP timeouts in 802.11 wireless networks. Though timeouts can be a problem for uploads from an 802.11 network, these timeouts are not spurious but are caused by a bottleneck at the access point. Once this bottleneck is removed, we find that spurious timeouts are rare, even in the face of large changes in numbers of active stations or PHY rate.

Construction of a PLC test bed for network and transport layer experiments

“…As expected, as flows are added to the system, the per-flow throughput is reduced by a factor of n. Interestingly, for 3 and 4 flows, timeouts occur for short bursts: this suggests that a similar TCP ACK starvation phenomenon to that reported by [16] may be an issue for HomePlug AV and IEEE 1901 systems as well. Future testing aims include verifying this result using a larger test set: as described in [16], the downlink node must transmit n/2 TCP ACKs for n uplink flows. While each source/uplink node has a single flow to buffer, the sink/downlink node has to deliver n/2 the number of channel accesses.…”

“…The distinction between the two tests are that for uplink tests, the access point receives multiple flows, while the downlink must source multiple flows. Potential problems can occur in the uplink for TCP flows if the ACKs cannot access the channel due to contention issues [16].…”

Buffer Sizing for 802.11-Based Networks

“…A consequence of using DCF is thus that when the number of upload flows increases, the uploads may produce enough TCP ACK packets to keep the AP's queue saturated. In fact, once there are two upload flows, TCP becomes unstable due to repeated timeouts (see [20] for a detailed demonstration), causing the unfairness issue discussed in Section II-C. Therefore, we present results for up to two uploads in Fig.…”