Lecture on Wireless Measurements & Modeling Prof. Maria Papadopouli University of Crete ICS-FORTH

1. Lecture on Wireless Measurements & Modeling Prof. Maria Papadopouli University of Crete ICS-FORTH http://www.ics.forth.gr/mobile

2. Agenda • Introduction on Mobile Computing & Wireless Networks • Wireless Networks - Physical Layer • IEEE 802.11 MAC • Wireless Network Measurements & Modeling • Location Sensing • Performance of VoIP over wireless networks • Mobile Peer-to-Peer computing

3. Empirical measurements • Can be beneficial in revealing • deficienciesof awireless technology • different phenomena of the wireless access & workload • Impel modelling efforts to produce more realisticmodels & synthetic traces • Enable meaningful performance analysis studiesusing such empirical, synthetic traces and models  Highlight the ability of empirical-based models to capture the characteristics of the user-workload and provide a flexible framework for using them in performance analysis

4. Propagation Models • One of the most difficult part of the radio channel design • Done in statistical fashion based on measurements made specifically for an intended communication system or spectrum allocation • Predicting the average signal strength at a given distance from the transmitter

5. Signal Power Decay with Distance • A signal traveling from one node to another experiences fast (multipath) fading, shadowing & path loss • Ideally,averaging RSS over sufficiently long time interval excludes the effects of multipath fading & shadowing  general path-loss model: P(d) = P0 – 10n log10 (d/do) n: path loss exponent P(d): the average received power in dB at distance d P0is the received power in dB at a short distance d0 _ _

6. Monitoring • Depending on type of conditions that need to be measured, monitoring needs to be performed at • Certain layers • Spatio-temporal granularities • Monitoring tools • Are not without flaws • Several issues arise when they are used in parallel for thousands devices of different types & manufacturers: • Fine-graindatasampling • Timesynchronization • Incompleteinformation • Dataconsistency

7. Monitoring & Data Collection • Finespatio-temporaldetail monitoring can  Improvetheaccuracyoftheperformanceestimates butalso • Increasetheenergyspendingsanddetectiondelay Networkinterfacesmay needto • Monitorthechannelinfiner& longertimescales • Exchangethisinformationwithotherdevices

8. Challenges in Monitoring (1/2) • Identificationof the dominant parameters through • sensitivity analysisstudies • Strategic placement of monitors at • Routers • APs, clients, and other devices • Automationof the monitoring process to reduce human intervention inmanaging the • Monitors • Collecting data

9. Challenges in Monitoring (2/2) • Aggregationofdatacollectedfromdistributedmonitorstoimprovetheaccuracywhilemaintaining low overhead in terms of • Communication • Energy • Cross-layermeasurements, collecteddataspanningfromthephysicallayeruptotheapplicationlayer, arerequired

10. Wireless Networks • Are extremely complex • Have been used for many different purposes • Have their own distinct characteristics due to radio propagation characteristics & mobility e.g., wireless channels can be highly asymmetric and time varying Note: Interaction of differentlayers & technologies creates many situations that cannot be foreseen during design & testing stages of technology development

11. Empirically-based Measurements • Real-life measurement studies can be particularly beneficial in revealing • deficienciesof awireless technology • different phenomena of the wireless accessand the workload • Rich sets of data can • Impel modeling efforts to produce more realisticmodels • Enable more meaningful performance analysis studies

12. Unrealistic Assumptions • Models & analysis of wired networks are valid for wireless networks • Wireless links are symmetric • Link conditions are static • The density of devices in an area is uniform • The communication pairs are fixed • Users move based on random-walk models

13. Wireless Access Parameters • Traffic workload • In different time-scales • In different spatial scales (e.g., AP, client, infrastructure) • In bytes, number of packets, number of flows, application-mix • Delays • Jitter and delay per flow • Statistics at an AP and/or channel • User mobility patterns • Link conditions • Network topology

14. Traffic Load Analysis • As the wireless user population increases, the characterization of traffic workloadcan facilitate • More efficient networkmanagement • Better utilization ofusers’ scarce resources • Application-based traffic characterization

15. Hourly Session arrival rates

16. Traffic Load at APs • Wide range of workloads that log-normality is prevalent • In general, traffic load is light, despite the long tails • No clear dependency with type of building the AP is locatedexists • Although some stochastic ordering is present in • Tail of thedistributions • Dichotomy among APs is prominent in both infrastructures:  APs dominated by uploaders APs dominated by downloaders • As the total received traffic atan AP↑ • There is also ↑in its total traffic sent • ↓ in the sent-to-received ratio

17. Traffic load at APs • Substantial number of non-unicast packets • Number of unicast received packets strongly correlated with number of unicast sent packets (in log-log scale) • Most of APs send & receive packetsof relatively small size • Significant number of APs show rather asymmetric packet sizes • APswith large sent & small receive packets • APs with small sent & large receive packets • Distributions of the number of associations & roaming operations are heavy-tailed • Correlation between the traffic load & number of associationsinlog-log scale

18. In general, the traffic load is light long tails RARE EVENTS OF LARGE SIZE

19. Wide-range of workloads • As total received traffic atan AP↑ •  its total traffic sent↑

20. APs with small amount of received traffic and large amount of sent traffic • As total received traffic atan AP↑ •  its total traffic sent/received ↓

21. The number of unicast received packets strongly correlatedwith the number of unicast sentpackets (in log-log scale)

22. Asymmetry in the sizes of sent & received packets at an AP Majority of APs with small sent and received packet sizes

23. Correlation between traffic load & # of associations

24. Application-based Traffic Characterization Using port numbers to classify flowsmay lead to significant amountsof misclassified trafficdue to: • Dynamic port usage • Overlapping port ranges • Traffic masquerading • Oftenpeer-to-peer & streaming applications: • Usedynamic ports to communicate • Port ranges of differentapplications may overlap • May try to masquerade their traffic under well-known “non-suspicious”ports, such as port 80

25. Desirable Properties for Models • Accuracy • Tractability • Scalability • Reusability • “Easy” interpretation

26. Related work • Rich literature in traffic characterization in wired networks • Willinger, Taqqu, Leland, Park on self-similarity of Ethernet LAN traffic • Crovela, Barford on Web traffic • Feldmann, Paxson on TCP • Paxson, Floyd on WAN • Jeffay, Hernandez-Campos, Smith on HTTP    • Traffic generators for wired traffic • Hernandez-Campos, Vahdat, Barford, Ammar, Pescape, … • P2P traffic • Saroiu, Sen, Gummadi, He, Leibowitz, … • On-line games • Pescape, Zander, Lang, Chen, … • Modelling of wireless traffic • Meng et al.

27. Dimensions in Modeling Wireless Access • Intended user demand • User mobility patterns • Arrival at APs • Roaming across APs • Channel conditions • Network topology

28. Mobility models • Group or individual mobility • Spontaneous or controlled • Pedestrian or vehicular • Known a priori or dynamic • Random-walk based models • Randway model in ns-2 • Markov-based model

29. A Very Simple Channel Model Gilbert model  Compute stationary probabilities Pib Idle Busy Pbb Pii Pbi • A channel can be in the idle or busy state • Markov-based model allows us to determine: • How much time the system spends in each state • Probability of being in a particular state •  In real rife, there is non-stationarity due dynamic phenomena

30. Main approaches for traffic generation • Packet-level replay • An exact reproduction of a collected trace in terms of packet arrival times, size, source, destination, content type  Reflects specific traffic conditions • Suffers from arbitrary delays e.g., interrupts, service mechanisms, scheduling processes  difficult to incorporate feedback-loop characteristics • Source-level generation  Allows the underlying network, protocol, & application layer to specify & control the packet arrival process • Simplest example: infinite source model

31. Our approach  Inspired by the source-level (or network independent) modelling Assumptions: • Client arrivals at an infrastructure (initiated by humans) at a large extent are not affected by the underlying network technology • Very low % of packet loss at the network layer  flow arrivals & sizes approximate intended user traffic demand

32. Internet disconnection Wired Network Router Switch AP3 Wireless Network User A AP 1 AP 2 Events User B Session 1 2 3 0 Flow Arrivals t1 t2 t3 t4 t5 t6 t7 time

33. Traffic Demand Parameters • Session • arrival process • starting AP • Flow within session • arrival process • number of flows • size (in bytes) Captures interaction between clients & network Above packet-level analysis

34. Wireless infrastructure & acquisition • 26,000 students, 3,000 faculty, 9,000 staff in over 729-acre campus • 488 APs (April 2005), 741 APs (April 2006) • SNMP data collected every 5 minutes • Several monthsof SNMP & SYSLOG data from all APs • Packet-headertraces: • Two weeks (in April 2005 and April 2006) • Captured on the link between UNC & rest of Internet via a high-precision monitoring card

35. Related modeling approaches • Flow-level modeling by Meng [mobicom ‘04] • No session concept • Weibull for flow interarrivals • Lognormalfor flow sizes • AP-level over hourly intervals • Hierarchical modeling by Papadopouli [wicon ‘06] Time-varying Poisson processfor session arrivals • BiPareto for in-session flow numbers & flow sizes • Lognormal for in-session flow interarrivals Sessions capture the non-stationarity of traffic workload

36. Modeling methodology • Selectionof models (e.g., various distributions) • Fittingparameters using empirical traces • Evaluationand comparison of models • Visual inspection e.g., CCDFs & QQ plots of models vs. empirical data • Statistical-based criteria e.g., QQ/simulation envelopes, Kullback-Lieblerdivergence • Systems-based criteria e.g., throughput, delay, jitter, queue size • Validation of models • Generalization of models

37. Synthetic trace generation

38. Synthetic traces based on empirical ones original data from the real-life infrastructure Produced by this process: Generate session arrivals within each session: generate number offlows for each flow: generate flow arrivals & sizes based on specific models • Session arrivals: using hourly, building-specific empiricaltraces • Flow-related data: usingempirical traces of different spatial scales

39. Model validation Use empirical data from different • tracing periods April 2005 & 2006 • spatial scales AP-level < building-level < building-type-level < network-wide • traffic conditions @ AP • campus-wide wireless infrastructures UNC, Dartmouth • Do the same distributions persist across these traces? Compare their performance (empirical traces: “ground truth”) YES!

40. Model evaluation Create synthetic data based on models Analysis with metrics not explicitlyaddressed by the models Statistical-based aggregate flow arrival count process aggregate flow interarrival (1st & 2nd order statistics) System-based: performance of an IEEE802.11 LAN traffic load and queuesize in various time scales per-flow & hourly aggregate throughput per-flow delay and jitter  Compare their performance (empirical traces: “ground truth”)

41. Modeling in Various Spatio-temporal Scales Objective Scales  Tradeoff with respect to accuracy, scalability & reusability

42. Scalability vs. Accuracy: Flow Interarrivals Spatial /Temporal Scales EMPIRICAL BDLG(DAY) BDLGTYPE(DAY) NETWORK(TRACE)

43. Scalability vs. Accuracy: Number of Flow Arrivals in an Hour BDLGTYPE(TRACE) BDLG(DAY) EMPIRICAL NETWORK(TRACE)

44. Model evaluation • Create synthetic data based on models • Analysis with metrics not explicitlyaddressed by the models • Statistical-based • aggregate flow arrival count process • aggregate flow interarrival (1st & 2nd order statistics) • System-based: performance of an IEEE802.11 LAN • traffic load and queuesize in various time scales • per-flow & hourly aggregate throughput • per-flow delay and jitter  Compare their performance (empirical traces: “ground truth”)  Dominant parameters ? Impact of application mix?

45. User C Statistical Analysis Monitoring & Data collection Collected traces Network monitor Internet Router Switch User A AP Wireless Networks AP Testbed Deployment User B Certain Protocol Evaluation at AP Network monitor Select testbed Protocol evaluation Cross validation Scenario Various traffic, channel conditions Iterative process

46. Simulation/Emulation testbed • TCP flows • UDP • Wired clients: senders • Wireless clients: receivers

47. Hourly aggregate throughput FLOW SIZE—FLOW (INTER)ARRIVAL EMPIRICAL Impact of flow size Fixed flow sizes & empirical flow arrivals (aggregate traffic as in EMPIRICAL) BIPARETO-LOGNORMAL-AP Pareto flow sizes, empirical flow arrivals BIPARETO-LOGNORMAL

48. Per-flow throughput FLOWSIZE—FLOWARRIVAL Pareto flow sizes & uniform flow arrivals BIPARETO-LOGNORMAL EMPIRICAL BIPARETO-LOGNORMAL-AP due to large % of small size flows (= MSS) Pareto flow sizes Fixed flow sizes & empirical number of flows

49. Impact of application mix on per-flow throughput TCP-based scenario AP with 85% web traffic AP with 80% p2p traffic AP with 50% web & 40% p2p traffic

50. UDP traffic scenario • Wireless hotspot AP • Wireless clients downloading • Wired traffic transmit at 25Kbps • Total aggregate traffic sent in CBR & in empirical is the same Empirical: 1.4 Kbps Bipareto-Lognormal-AP: 2.4 Kbps Bipareto-Lognormal: 2.6 Kbps Large differences in the distributions