Understanding Link-Layer Behaviorin Highly Congested IEEE 802.11b Wireless NetworksAmit P. Jardosh, Krishna N. Ramachandran, Kevin C. Almeroth, Elizabeth M. Belding-RoyerDepartment of Computer ScienceUniversity of California, Santa Barbara{amitj,krishna,almeroth,ebelding}@cs.ucsb.eduAbstractThe growing deployment and concomitant rise in wireless networkusage necessitates the comprehensive understanding of its behav-ior. More importantly, as networks grow in size and number ofusers, congestion in the wireless portion of the network is likelyto increase. We believe there is a strong need to understand theintricacies of the wireless portion of a congested network by inter-preting information collected from the network. Congestion in awireless network can be best analyzed by studying the transmis-sion of frames at the link layer. To this end, we use vicinity sniffingtechniques to analyze the link layer in an operational IEEE 802.11bwireless network. In this paper, we discuss how congestion in anetwork can be estimated using point-to-point link reliability. Wethen show how link reliability is correlated with the behavior oflink-layer properties such as frame retransmissions, frame sizes,and data rates. Based on the results from these correlations, ourhypothesis is that the performance of the link layer in congestednetworks can be improved by (1) sending smaller frames, and/or(2) using higher data rates with a fewer number of frames sent.Categories and Subject Descriptors: C.2.3 [Network Operations]:Network Monitoring, C.4 [Performance of Systems]: MeasurementTechniques and Performance Attributes.General Terms: Measurement, Performance.Keywords: IEEE 802.11b, Network Congestion, Performance Ana-lysis.1. INTRODUCTIONThe growing deployment and concomitant rise in wireless net-work usage necessitates the comprehensive understanding of itsbehavior. To address this issue, researchers have computed mod-els and statistics for aggregated traffic flow patterns [2, 5, 7, 8,10], movement patterns [3, 4, 11], and wireless link quality mea-sures [1]. In most cases, researchers have utilized: (1) aggregateflow analysis techniques to gain insight into transport and applica-tion layer metrics on the wired portion of the network, and/or (2)SNMP and syslog information from access points to model move-ment and association patterns.Permission to make digital or hard copies of all or part of this work forpersonal or classroom use is granted without fee provided that copies arenot made or distributed for profit or commercial advantage and that copiesbear this notice and the full citation on the first page. To copy otherwise, torepublish, to post on servers or to redistribute to lists, requires prior specificpermission and/or a fee.SIGCOMM’05 Workshops, August 22–26, 2005, Philadelphia, PA, USA.Copyright 2005 ACM 1-59593-026-4/05/0008 ...$5.00.Although previous work has contributed to the understanding ofwireless networks, it is the RF (radio frequency) environment that isusually responsible for performance bottlenecks. The wired back-haul network is typically sufficiently provisioned to handle largetraffic flows to and from the wireless network.Recent work has observed the performance of and detected ano-malies in the MAC layer through measurement experiments in ei-ther controlled environments [12] or on small-scale wireless net-works in environments free of ancillary interference [9]. However,the results derived from small-scale and controlled experiments arenot representative of the variations in MAC layer properties thatare axiomatic to large-scale congested networks. The study ofMAC layer properties provides a good measure of the contentionfor channel occupancy in the wireless medium which can be usedto estimate congestion in the network.As wireless networks become densely-populated and heavily-utilized, there arises a need to fully understand the intricacies ofthe wireless portion of these networks by interpreting collectedMAC layer information. A large-scale, congested wireless networkis characterized by extensive medium occupancy, high throughput,frequent bit errors, numerous retransmissions, and significant datarate variations. To fully understand the interplay of these proper-ties, it is essential to monitor and study an operational congestednetwork.To this end, we analyzed a wireless network deployed at a re-cently concluded 62ndInternet Engineering Task Force (IETF)1meeting held in Minneapolis, MN. The meeting was held March6–11, 2005 and was attended by 1138 participants. Almost all ofthe participants used laptops or other wireless devices. The wire-less network at the meeting consisted of 38 IEEE 802.11b accesspoints deployed on three adjacent floors of the venue. An analysisof the data shows that the large number of participants and accesspoints resulted in heavy utilization of the wireless network alongwith multiple periods of congestion.The method adopted to collect information from the wirelessportion of the IETF network is referred to as wireless network mon-itoring [12] or vicinity sniffing. In order to capture network framesthat were transmitted by access points and user devices in the net-work, we chose to place three laptops (wireless network sniffers)at a single vantage point in the network. The location was strate-gically selected so that our sniffers could overhear and record themaximum possible information.One of the most interesting periods of network congestion duringthe week-long meeting was observed during a plenary session heldbetween 19:30 hrs and 22:30 hrs on March 10, 2005. The sessionhad a total of over 500 attendees in an open meeting room of size 30× 25 meters. The data collected during this session shows that (1)1http://www.ietf.org/11at any given time during the meeting, there were 250 to 300 usersout of the total attendees associated with the network on channels1, 6, and 11; (2) the participants sent and received a total of over 9million data frames; (3) the most active AP, as recorded in our dataset, sent and received a total of 1.37 million data frames; and (4)the network throughput peaked at 5.5 Mbps, close to the theoreticaloptimum [6].In this paper we focus on this plenary session. The network con-ditions at the plenary best satisfy both of our study criteria of beinglarge and congested. Through analysis of the data, we make thefollowing hypotheses: (1) as congestion in the network intensifies,smaller frames are more likely to be successfully received;