Characterizing Usage in Wireless Networks: OutlineUnderstand user’s behavior mobility Utilization of APs Applications used in wireless networks www, VoIP, p2p application, streaming media Wireless devices usedCharacterizing wireless use is important Providers: provision the network Designers: standard features for high-throughput Software developers: wireless-aware applications Changes in usage is expected New wireless devices: PDAs, printers, audio players New applications: real-time multimediaThe Changing Usage of a Mature Campus-wide Wireless NetworkCampus-wide usage studyExtensive data set collectionsData decomposition and analysisClient usage trends from 2001 to 2004AP utilization trendsTraffic trends from 2001 to 2004Application changesSpecific applications: VoIP trendsSpecific applications: P2P trendsSpecific applications: streaming trendsMobility trends: Users are not very mobileCharacterizing Mobility and Network Usage in a Corporate Wireless Local-Area NetworkGoal of the paperExperiment DataPopulation characteristicsUtilization of APsUsers and loadMobility characteristicsPrevalenceSlide 24PersistenceSummary1Characterizing Usage in Wireless Networks: OutlineIntroductionMotivationCampus-wide usage studyCorporate mobility and usage studySummary2Understand user’s behaviormobilityUtilization of APsApplications used in wireless networkswww, VoIP, p2p application, streaming mediaWireless devices usedCharacterizing Usage in Wireless Networks3Characterizing wireless use is importantProviders: provision the network Designers: standard features for high-throughputSoftware developers: wireless-aware applicationsChanges in usage is expectedNew wireless devices: PDAs, printers, audio playersNew applications: real-time multimediaWhy important?4The Changing Usage of a Mature Campus-wide Wireless NetworkTristan Henderson, David Kotz, Ilya AbyzovDartmouth College5Campus-wide usage studyExtensive data collection at Dartmouth college over 17 weeksNovember 2003 – February 2004190 buildings on 200 acres5500 students / 1200 faculty•3200 – 3300 undergraduate students•required to own a computer (97% laptops)6Extensive data set collections4 sources of data consisting of over32 million syslog messages (1 sec resolution)16 million SNMP polls (5 min interval)4.6 TB of packets sniffed5.16 GB Call detail records (CDRs) for VoIP7Data decomposition and analysisMain goal: Understand user's behavior8Client usage trends from 2001 to 2004Behaviors that remain the sameUsage is still diurnalSame proportion of heavy users Same busiest buildingsBehaviors that have changed# of cards increased linearlyRoaming increasedAP utilization increased9AP utilization trendsFall/Winter 2003/4Fall 2001Avg = 66.4%Avg = 76.4%Approx 2.5x increase10Traffic trends from 2001 to 2004Traffic behavior changesOverall traffic increased by approximately 3xApplications changedDestination reversal (now more on-campus traffic)Traffic behavior constantsResidences generate the most traffic11Application changesReported proportionswww decreased from 62.9% to 28.6%P2P increased from 5.2% to 19.3%filesystems increased from 5.3% to 21.5%streaming increased from 0.9% to 4.6%Fall/Winter 2003/4Fall 200112Specific applications: VoIP trendsVoIP usage behaviorsUsage is diurnalNumber of devices does not grow muchUsers made short callsWireless users made few callsmedian call: 42 secmedian wireless: 31 sec13Specific applications: P2P trendsFiles were downloaded and uploadedTraffic was predominantly internal (72.7%)Few users responsible for most traffic147 cards (2%) saw over 1MB traffic10 cards (6.8% of the 2%) saw over 50% of recorded traffic14Specific applications: streaming trendsMost streaming was inboundOutbound itunes traffic represent music sharingMost traffic was within campus (79.6%)Streaming for teaching purposes produces large files15Mobility trends: Users are not very mobile50% of users spent 98.7% of time at homeespecially in residential, academic, and administration buildingsHigher than Castro paper50% of users spent nearly 10 minat single APLonger than Castro paper16Characterizing Mobility and Network Usage in a Corporate Wireless Local-Area NetworkMagdalena Balazinska – MITPaul Castro – IBM Research17Goal of the paperPrevious studies analyze usage of wireless networks in:University Building (Tang and Baker)University Campus (Kotz and Essien)Large Auditorium (Balachandran et al.)Metropolitan area network (Tang and Baker)Evaluate results from data collected in a Corporate Wireless Local-area NetworkFocus: Population characteristics, load distribution across APs, users level of activity, user mobility18Experiment DataThree Corporate buildingsOne Large (LBldg): 131 AP, ~10 miles away from the other two buildingsOne Medium (MBldg): 36 APOne Small (SBldg): 10 APMBldg and SBldg are adjacentTopology: 1 AP per corridor, extra in highly used rooms (i.e. customer lab in SBldg)Number of users = Number of MAC addressesData collected every 5 minutes, for 30 days, using SNMP19Population characteristics1366 distinct users seen: 796 mainly in LBldg, 437 in MBldg, 133 in SBldgSimilar results to working campus locations, different from dormitories or metropolitan networksPatterns in the population reflect patterns in n. of users20Utilization of APsGood coverage implies partial usage of APs21Users and loadAP throughput: amount of bytes that an access point forwards for any user in either direction in a given period of timeNumber of users and load of a specific AP are weakly dependent22Mobility characteristicsFew users visit all three buildingsDifferent mobility among usersMost of the time spent at one (home) locationUsers have higher activity on guest locations23PrevalencePrevalence: amount of time a user spends with a given APPrevalence matrixPrevalence distribution24Prevalence5 categories of users: Highly mobile, Somewhat Mobile, Regular, Occasionally Mobile, Stationary25PersistencePersistence: amount of time a user that a user stay associated w/ an AP before moving to another AP26SummaryUsage characterizationNumber of wireless cards; utilization of APsSpecific applications•VoIP, P2P, streaming mediaMobility•Prevalence, persistenceWork is
View Full Document