Deflating link buffers in a wireless mesh network

Abstract

We analyze the problem of buffer sizing for backlogged TCP flows in 802.11-based wireless mesh networks. Our objective is to maintain high network utilization while providing low queueing delays. Unlike wired networks where a single link buffer feeds a bottleneck link, the radio spectral resource in a mesh network is shared among a set of contending mesh routers. We account for this by formulating the buffer size problem as sizing a collective buffer distributed over a set of interfering nodes. In this paper we propose mechanisms for sizing and distributing this collective buffer among the mesh nodes constituting the network bottleneck. Our mechanism factors in the network topology and wireless link rates, improving on pre-set buffer allocations that cannot optimally work across the range of configurations achievable with 802.11 radios. We evaluate our mechanisms using simulations as well as experiments on a testbed. Our results show that we can reduce the RTT of a flow by 6× or more, at the cost of less than 10% drop in end-to-end flow throughput.

TCP performance evaluation over backpressure-based routing strategies for wireless mesh backhaul in LTE networks

“…The RTT abstracts network latency and is defined as the time taken by a TCP segment to reach the UE plus the time taken by the server to receive the correspondent ACK from the UE. This parameter is gathered from the TCP data sender each time it receives a non-duplicated ACK segment: therefore, retransmitted or out-of-order data packets are not taken into EPC is connected to the Internet through another PTP wired link (characterized by a bandwidth of 10 Gb/s and a propagation delay of 5 ms); the queue sizes of the backhaul network are set to the 100 % of each link bandwidth-delay product, making an effort to reduce default buffer size, as suggested in [8]. Regarding the Radio Access Network (RAN), we used European frequencies and a Okumura-Hata propagation model [9]: the LTE connection between UE and the eNodeB is modeled inside an urban environment of a medium city.…”

WQM

“…Distributed Neighborhood Buffer [6] targets the buffer sizing problem in Wireless Mesh Networks (WMNs). The authors first calculate a cumulative neighborhood buffer based on link interference constraints, and then distribute it among competing nodes using a cost function to ensure efficient spectral utilization.…”

Reduction of Buffering Requirements: Another Advantage of Cooperative Transmission

“…In this letter, We analyze this cost and contrast the cooperative case to the non-cooperative case. Existing work in this domain only aims at developing buffer overflow management mechanisms (see [2]). Up to the best of our knowledge, analyzing the buffering cost in wireless sensor networks from an end-to-end perspective is quite novel.…”