“Sociological Orbits” Mobility Profiling and Routing for Mobile Wireless Networks

Lab for Advanced Network Design, Evaluation and Research

“Sociological Orbits”Mobility Profiling and Routingfor Mobile Wireless Networks

Joy GhoshPh.D. Dissertation Defense

Major Advisor: Dr. Chunming Qiao


Outline Mobility - Impact on Routing / Advantages Acquaintance Based Soft Location Management

(ABSoLoM) Sociological ORBIT Mobility Framework Mobility Profiling Techniques and Applications Sociological Orbit aware Location Approximation and

Routing (SOLAR) – MANET & ICMAN Theoretical Analysis of SOLAR

Routing problem formulation for ICMAN Approximation algorithm for delivery probability Mathematical model for computing contact probability Edge-constrained routing protocol and its performance

Concluding Remarks


The Overall Picture


Mobility Impact on Routing

Node Mobility Dynamic network topology Proactive protocols are inefficient

Need to exchange control packets too often Leads to congestion E.g., Distance Vector, Link State

Reactive protocols are better suited, but Locating a node incurs more delay Route maintenance is tricky as nodes move E.g., Dynamic Source Routing (DSR), Location Aided

Routing (LAR)


Framework for analyzing impact of mobility on protocol performance F. Bai, N. Sadagopan,

and A. Helmy, “Important: a framework to systematically analyze the impact of mobility on performance of routing protocols for adhoc networks”, Proceedings of IEEE INFOCOM '03, vol. 2, pp. 825-835, March 2003.


Greedy Geographic Forwarding

Pros Less affected by mobility than source routes Smaller header size (no path cached)

Cons Nodes need to know own location Needs sufficient node density

Workarounds for local maxima Broadcast Planar graph perimeter routing (e.g., GPSR)


Advantages of Node Mobility – Individual node’s view of network


Advantages of Node Mobility – Node’s view of network through “acquaintances”


Acquaintance Based Soft Location Management (ABSoLoM) Forming and maintaining acquaintances Limit number of acquaintances Keep updating acquaintances of location Query acquaintances for destination location Limit query propagation by logical hops On learning of destination, use geographic

forwarding to send packets to destination Nosy Neighbors

Can respond to query if destination’s location is known Caches node locations while forwarding certain packets


Performance Analysis

Simulated in GloMoSim LAR & DSR borrowed from the GloMoSim distribution Implementation of SLALoM by Dr. Sumesh J. Philip (author) ABSoLoM parameters

Number of friends = 3 Maximum logical hops = 2

100 nodes in 2000m x 1000m for 1000s Random Waypoint mobility

Velocity = 0m/s-10m/s; Pause = 15s Random CBR connections varied in simulation

50 packets per connection; 1024 bytes per packet


Results – I.a: Throughput vs. Load


Results – I.b: Overhead vs. Load


Simulation Results – II (a) Hop Latency vs. Load & (b)

Throughput vs. Mobility


Parallel growth of models and protocols Practical mobility models

Random Waypoint simple, but impractical!! Entity based individual node movement Group based collective group movement Scenario based geographical constraints

Mobility pattern aware routing protocols Mobility tracking and prediction Link break estimation Choice of next hop


Our Motivation Not to suggest only a practical mobility model MANET is comprised of wireless devices carried

by people living within societies Society imposes constraints on user movements Study the social influence on user mobility Realization of special regions of some social

value Identify a macro level mobility profile per user Use this profile to aid macro level soft location

management and routing


Mobile Users

• influenced by social routines

• visit a few “hubs” /

places (outdoor/indoor) regularly

• “orbit” around (fine to coarse grained) hubs at several levels

Sociological Orbit Framework


Illustration of A Random Orbit Model

(Random Waypoint + Corridor Path)Conference Track 1

Conference Track 3

Cafeteria

Lounge

Conference Track 2

Conference Track 4

PostersRegistration

Exhibits


Random Orbit Model


Traces Used Profiling techniques applied to ETH Zurich traces

Duration of 1 year from 4/1/04 till 3/31/05 13,620 wireless users, 391 APs, 43 buildings Grouped users into 6 groups based on degree of activity Selected one sample (most active) user from each group

Mapped APs into buildings based on AP’s coordinates, and each building becomes a “hub” Converted AP-based traces into hub-based traces

Other traces Expect similar results from Dartmouth’s traces No sufficient AP location info from other traces UMass’s traces are for buses, more predictable than users Need to obtain actual users’ traces with GPS


Hub-centric Parameters - I


Hub-centric Parameters - II


Hub Based Mobility Profiles and Prediction On any given day, a user may regularly visit a small number of “hubs”

(e.g., locations A and B) Each mobility profile is a weighted list of hubs, where weight = hub visit

probability (e.g., 70% A and 50% B) In any given period (e.g., week), a user may follow a few such “mobility

profiles” (e.g., P1 and P2) Each profile is in turn associated with a (daily) probability (e.g., 60% P1

and 40% P2) Example: P1={A=0.7, B=0.5} and P2={B=0.9, C=0.6}

On an ordinary day, a user may go to locations A, B and C with the following probabilities, resp.: 0.42 (=0.6x0.7), 0.66 (= 0.6x0.5 + 0.4+0.9) and 0.24 (=0.4x0.6)

20% more accurate than simple visit-frequency based prediction Knowing exactly which profile a user will follow on a given day can result in

even more accurate prediction

On any given day, a user may regularly visit a small number of “hubs” (e.g., locations A and B)Each mobility profile is a weighted list of hubs,

where weight = hub visit probability (e.g., 70% A and 50% B)

In any given period (e.g., week), a user may follow a few such “mobility profiles”

(e.g., P1 and P2)Each profile is in turn associated with a (daily) probability (e.g., 60% P1 and 40% P2) Example: P1={A=0.7, B=0.5} and P2={B=0.9, C=0.6}On an ordinary day, a user may go to locations A, B & C with the following probabilities: 0.42 (=0.6x0.7), 0.66 (= 0.6x0.5 + 0.4+0.9), 0.24 (=0.4x0.6)• 20% more accurate than simple visit-frequency based prediction• Knowing exactly which profile a user will follow on a given day can result in even more accurate prediction


Orbital Mobility Profiling Obtain each user’s daily hub lists as binary vectors Represent each hub list (binary vector) as a point in

a n-dimensional space (n = total number of hubs) Cluster these points into multiple clusters, each with

a mean Using the Expectation-Maximization (EM) algorithm based

on a Mixture of Bernoulli’s distribution Probe other classification methods: Bayesian-Bernoulli’s

Each cluster mean represents a mobility profile, described as a probabilistic hub visitation list

User’s mobility is aptly modeled using a mixture of mobility profiles with certain “mixing proportions”


Profiling illustration

Obtain daily hub stay durations

Translate to binary hub visitation vectors

Apply clustering algorithm to find mixture of profiles


Profile parameters for all sample users


Hub-based Location Predictions - I Unconditional Hub-visit Prediction

Prediction Error = Incorrect hubs predicted over Total hubs SPE – Statistical based Prediction Error

SPE-ALL: (n+1)th day prediction based on hub-visit frequency from day 1 through day n

SPE-W7 : (n+1)th day prediction based on hub-visit frequency within last week, i.e., day (n-7) through day n

PPE – Profile based Prediction Error PPE-W7 : (n+1)th day prediction based on profiles of the last

week, i.e., day (n-7) through day n Prediction Improvement Ration (PIR)

PIR-ALL = (SPE-ALL – PPE-W7) / SPE-ALL PIR-W7 = (SPE-W7 – PPE-W7) / SPE-W7


Unconditional Prediction Results

The profile mixing proportions vary with every window of n days


Hub-based Location Predictions - II Conditional Hub-visit Prediction

Improvement given current profile is known/identifiable It is possible sometimes to infer profile from current hub

information alone Our method effectively leverages information when available

Sample user categoriesTarget Hub ID: will the user visit this hub?The current day in questionPredicted probability using visit frequency Indicator (Current) HubCurrent ProfilePredicted probability based on profileActually visited Ht on day D or not


Hub-based Location Predictions - III Hub sequence prediction based on hub transitional probability

Prediction Accuracy = 1 – (incorrect predictions / total predictions) Scenario 1: only starting hub is known for sequence prediction Scenario 2: hub prediction is corrected at every hub in sequence Better performance with increasing knowledge – intuitive

Statistical based Prediction Accuracy (SPA) – no profile informationProfile based Prediction Accuracy (PPA) – no time informationTime based Prediction Accuracy (TPA) – temporal profiles


Applications of Orbital Mobility Profiles Location Predictions and Routing within MANET and ICMAN

Anomaly based intrusion detection unexpected movement (in time or space) sets off an alarm

Customizable traffic alerts alert only the individuals who might be affected by a specific traffic condition

Targeted inspection examine only the persons who have routinely visited specific regions

Environmental/health monitoring identify travelers who can relay data sensed at remote locations with no APs


Profile based Routing within MANET

Build a sociological orbit based mobility model (Random Orbit)

Assume that mobility profiles are obtained Devise routing protocols to leverage mobility

information within MANET setting Key assumption – geographical forwarding is

feasible


Sociological Orbit aware Location Approximation and Routing (SOLAR) - Basic Every node knows

Own coordinates, Own Hub list, All Hub coordinates Periodically broadcasts Hello

SOLAR-1 : own location & Hub list SOLAR-2 : own location & Hub list + 1-hop neighbor Hub lists

Cache neighbor’s Hello Build a distributed database of acquaintance’s Hub lists

Unlike “acquaintanceship” in ABSoLoM, SOLAR has No formal acquaintanceship request/response its not mutual Hub lists are valid longer than exact locations lesser updates

For unknown destination, query acquaintances for destination’s Hub list (instead of destination’s location), in a process similar to ABSoLoM


Sociological Orbit aware Location Approximation and Routing (SOLAR) - Advanced Subset of acquaintances to query

Problem: Lots of acquaintances lot of query overhead Solution: Query a subset such that all the Hubs that a node learns of from its

acquaintances are covered Packet Transmission to a Hub List

All packets (query, response, data, update) are sent to node’s Hub list To send a packet to a Hub, geographically forward to Hub’s center If “current Hub” is known – unicast packet to current Hub Default – simulcast separate copies to each Hub in list

We compared simulcast, unicast, multicast – simulcast had best performance with higher cost of overhead and delay

On reaching Hub, do Hub local flooding if necessary Improved Data Accessibility – Cache data packets within Hub

Data Connection Maintenance Two ends of active session keep each other informed Such location updates generate “current Hub” information


Sociological Orbit aware Location Approximation and Routing (SOLAR) –

IllustrationHub A

Hub B

Hub C

Hub D

Hub E

Hub I

Hub FHub G

Hub H


Performance Analysis Metrics

Data Throughput (%) Data packets received / Data packets generated

Relative Control Overhead (bytes) Control bytes send / Data packets received

Approximation Factor for E2E Delay Observed delay / Ideal delay To address “fairness” issues!


Performance Analysis Parameters


Results – I.a : Throughput vs. Hubs


Results – I.b : Overhead vs. Hubs


Results – I.c : Delay vs. Hubs


Routing challenges in ICMAN

ICMAN Features of DTN/ICN + MANET Lack of infrastructure and any central control May not have an end-to-end path from source to

destination at any given point in time Conventional MANET routing strategies fail User mobility may not be deterministic or

controllable Devices are constrained by power, memory, etc. Applications need to be delay/disruption tolerant


User level routing strategies Deliver packets to the destination itself Intermediate users store-carry-forward the packets Mobility profiles used to compute pair wise user contact probability

P(u,v) via Semi-Markov Process Form weighted graph G with edge weights w(u,v) = log (1/P(u,v)) Apply modified Dijkstra’s on G to obtain k-shortest paths (KSP) with

corresponding Delivery probability under following constraints Paths are chosen in increasing order of total weights (i.e., minimum first) Each path must have different next hop from source

S-SOLAR-KSP (static) protocol Source only stores set of unique next-hops on its KSP Forwards only to max k users of the chosen set that come within radio range

within time T D-SOLAR-KSP (dynamic) protocol

Source always considers the current set of neighbors Forwards to max k users with higher delivery probability to destination


Hub level routing strategy Deliver packets to the hubs visited by destination Intermediate users store-carry-forward the packets Packet stored in a hub by other users staying in

that hub (or using a fixed hub storage device if any) Mobility profiles used to obtain delivery probabilities

(DP), not the visit probability, of a user to a given hub i.e. user may either directly deliver to hub by traversing to

the hub, or may pass onto other users who can deliver to the hub

Fractional data delivered to each hub proportional to the probability of finding the destination in it

Routing Strategy SOLAR-HUB protocol


SOLAR-HUB Protocol Pd

nihj: delivery probability (DP) of user ni to hub hj

Ptnihj: probability of user ni to travel to hub hj

h(ni): hub that user ni is going to visit next Pc

nink(hj): probability of contact between users ni & nj in hub hj

N(ni): neighbors of user ni

Pdnihj = max(Pt

nihj, maxk(Pcnink(h(ni))*Pt

nkhj)) Source ns will pick ni as next hop to hub hj as:

{ni | max(Pdnihj), ni Є N(ns)} iff P

dnihj > Pd

nshj

Packet Delivery Scheme Source transmits up to k copies of message

k/2 to neighbors with higher DP to “most visited” hub k/2 to neighbors with higher DP to “2nd most visited” hub

Downstream users forward up to k users with higher DP to the hub chosen by upstream node


Simulation Parameters for GloMoSim


Performance – Number of Hubs

• Overhead of EPIDEMIC is much more than others and had to be omitted from plot

• Overall D-SOLAR-KSP performs best


Performance – Number of Users

• Overhead of EPIDEMIC is much more than others and had to be omitted from plot

• Overall D-SOLAR-KSP performs best like before because it is the most opportunistic in forwarding to any of its current neighbors


Performance – Cache Size (Only SOLAR)

• All versions fair better with more cache



Performance – Cache Timeout (Only SOLAR)

• All versions fair better with larger timeout



Routing problem in probabilistic graphs Objective: maximize delivery probability from nodes s to t under

various constraints G = (V,E) be a complete directed graph

V = ICMAN users; E = probabilistic contact between users Let A be a routing algorithm and G(A) be the delivery sub-graph

induced by A Delivery probability is then s,t-connectedness probability (two-

terminal reliability) denoted by Conn2(G(A)) Goal is to find a delivery sub-graph G(A) to maximize Conn2(G(A))

we have shown it to be #P-hard 2 Possible approaches

Approximate Conn2(G(A)) by another polynomial time function Develop heuristics for A for which Conn2(G(A)) can be

approximated in polynomial time


Approximation algorithm G = (V, E) where edge probability between nodes

u and v is pe(u,v) (a) In G, starting from s, all nodes choose at most k

downstream edges to get Gk = (V, Ek) (b) Weight of each edge in Gk is set to

we(u,v) = -1 * log (pe(u,v)) to get G’k say Compute shortest path from s to all nodes in G’k to

get Gsp = (V, Esp) & assign BFS level #s (c) Reset we(u,v) = pe(u,v) & add all edges (v,d) that

were in G to get G’ = (V, E’) (d) Let Pd(u,v) be delivery probability of node u to v Apply Algorithm 1 to G’ to get Pd(s,d)

Start with any u ≠ d with maximum level # Pd(u,d) = 1 – Πk

1(1 – pi) Where pi = we(u,vi) * Pd(vi, d) for all edges (u,vi)


Algorithms for delivery probability

Calculate all paths from s to d Apply Algorithm 2 by rules of

inclusion and exclusion


Performance comparison of approximation algorithm with optimal


Contact Probability using a Semi-Markov Chain Hub transitional probability of user X from hubs h to h’ = βX

hh’>0, Σh≠h’βXhh’=1

Inter-hub transition time exponential with mean λXhh’

Xt be the hub X is in at time t Et be the hub stay time at Xt-1 before coming to Xt

distributed as power law with exponent λXh

This movement can be modeled with a Semi-Markov Chain (SMC) State space of X: Ix = S U { (h, h’) | h,h’ Є S, h ≠ h’}

Where, S = set of X’s hubs, (h, h’) = movement of X from h to h’ Holding times at states in S are power law distributed Holding times at states in (h, h’) are exponentially distributed

State transitional probability pXij

= βXhh’ when i = h and j = (h,h’)

= 1 when i = (h,h’) and j = h’ = 0 otherwise


Contact Probability using a Semi-Markov Chain We consider similar formulation for user Y with hub set T

Let R = S ∩ T ≠ 0 Objective

Find probability of X meeting Y at time t (~equilibrium) Find probability of X meeting Y at a particular hub h Є R, at time t

Combined SMC: {Zt | t ≥ 0} Cartesian product of SMCs of X and Y State space I = IX x IY; states (x, y) x Є IX, y Є IY

Sojourn times at x and y are either exponential, or power law with known parameters Sojourn time at (x, y) may be calculated with simple exercises

Jumping probabilities If sojourn time Ti at state i of X < sojourn time Ti’ at state i’ of Y

pXY(i,i’)(j,i’) = pX

ij

If sojourn time Ti at state i of X > sojourn time Ti’ at state i’ of Y pXY

(i,i’)(i,j’) = pYij

EMC of Z is ergodic as long as EMC of X and Y are ergodic Find only occupancy probabilities πXY

(h, h) at equilibrium for state (h, h), hЄR Probability that X meets Y at equilibrium ΣhЄRπXY

(h, h)


Edge-constrained routing – EC-SOLAR-KSP

EC-SOLAR-KSP1 L = |E| EC-SOLAR-KSP2 L = 0.8 * |E| EC-SOLAR-KSP3 L = 0.6 * |E|


Concluding Remarks - Contributions Use of acquaintances for soft location management

Sociological ORBIT framework and mobility models

Profiling user mobility and predicting locations

Using mobility profiles for routing within MANET and ICMAN

Formulation and analysis of a novel routing problem within probabilistic graphs


Concluding Remarks – Future work More efficient profiling techniques

Overcome shortcomings – bias towards hub visits Use other tools like time series analysis

Profile exchange and management Profile lifetime in cache Distribution of profile to minimize query radii

Solutions to our routing optimization problem Develop an optimal routing algorithm that gives a delivery

sub-graph which maximizes the delivery probability


Related Publications Journal

Joy Ghosh, Sumesh J. Philip, Chunming Qiao, "Sociological Orbit aware Location Approximation and Routing (SOLAR) in MANET" - Accepted for publication in ELSEVIER Ad Hoc Networks Journal, Nov 2005

Workshops Joy Ghosh, Hung Q. Ngo, Chunming Qiao, "Mobility Profile based Routing within Intermittently Connected

Mobile Ad hoc Networks (ICMAN)" - Accepted for publication in IWCMC 2006 Delay Tolerant Mobile Networks workshop, Vancouver, Canada, July 2006

Joy Ghosh, Matthew J. Beal, Hung Q. Ngo, Chunming Qiao, "On Profiling Mobility and Predicting Locations of Wireless Users" - Accepted for publication in ACM/SIGMOBILE REALMAN 2006 workshop at ACM Mobihoc '06, Florence, Italy, May 2006

Conferences Joy Ghosh, Cedric Westphal, Hung Ngo, Chunming Qiao, "Bridging Intermittently Connected Mobile Ad hoc

Networks (ICMAN) with Sociological Orbits" - Poster at INFOCOM '06, Barcelona, Spain 2006 (April)

Joy Ghosh, Sumesh J. Philip, Chunming Qiao, "Sociological Orbit aware Location Approximation and Routing in MANET" - Proceedings of IEEE Broadnets, Boston, MA, 2005 (October)

Joy Ghosh , Sumesh J. Philip, Chunming Qiao, "Poster Abstract: Sociological Orbit aware Location Approximation and Routing (SOLAR) in MANET" - Poster at ACM International Symposium on Mobile Ad Hoc Networking and Computing, MobiHoc 2005 (May)

Joy Ghosh, Sumesh J. Philip, Chunming Qiao, "Acquaintance Based Soft Location Management (ABSLM) in MANET" - Proceedings of IEEE Wireless Communications and Networking Conference 2004 (March)

Technical Reports http://www.cse.buffalo.edu/~joyghosh/solar.html


Thank You!

Questions?


Source Routing (DSR, LAR)Return


Geographic Forwarding may help

(nodes must know own location)Return


Send request to all All pending acquaintances Few accepted request Time0: some nodes move out Time1: timeout terminates

acquaintance Time2: some move back in Time3: some move out again Time4: timeout terminates

acquaintance

Forming & maintaining acquaintances

Non Acqntnce Pending Acqntnce Accepted Acqntnce

Return


Querying AcquaintancesReturn


Fairness in Delay Comparison

return


Expectation-Maximization (EM)

Hub lists

y(1) = “110011”

y(2) = “110000”

y(3) = “000011”

y(4) = “101010”

y(5) = “010101”

Daily hub list Cluster mean

2-D example view

Initializations

Weighted means

ρ(1) = “0.7, 0.8, 0.2, 0.3, 0.7, 0.7”

ρ(2) = “0.1, 0.3, 0.9, 0.8, 0.3, 0.2”

Mixing proportions

π = {π1, π2} = {0.5, 0.5}

r(i)j C1 C2

y(1) 0.9 0.1

y(2) 0.6 0.4

y(3) 0.6 0.4

y(4) 0.5 0.5

… … …

ρ(1) = “0.9, 0.9, 0.1, 0.1, 0.9, 0.9”

ρ(2) = “0.1, 0.1, 0.9, 0.9, 0.1, 0.1”

π = {π1, π2} = {0.7, 0.3}

return


Problem complexity: #P-hard ! Valiant proved Conn2(G) to be #P-complete in 1979; we reduce it to our problem

In directed graph D = (V,E) let pij be all edge probabilities with source s and destination t; let c be the LCM of all denominators of pij (c is polynomial in input size)

If we have a procedure to compute Conn2(G) ≤ c’/c for any c’ ≤ c, we can compute Conn2(G) by simple binary search

Our objective: find routing algorithm A, which finds delivery sub-graph D = G(A) to maximize Conn2(D) – a solution to this can be used to decide if Conn2(G) ≤ c’/c !!

Add path with k edges to D to get G with Πki=1pi = c’/c + ε, where ε < 1/c

Our aim: find sub-graph H of G with |E(H)| ≤ k (edge

constraint) Routing algorithm A returns

Upper part Conn2(D) ≤ c’/c

Lower part Conn2(D) > c’/c

“A” can be used to decide if Conn2(G) ≤ c’/c

“A” is at least as hard as Conn2(G)

return


Acquaintance Ai has a Hub list Hi = {h1, h2, …, hm} where hi is a Hub

H = {H1, H2, …, Hn} is the set of Hub lists covered by A1, A2, …, An

C = H1 U H2 U … U Hn is the set of all Hubs covered by A1, A2, …, An Objective: find a minimum subset

This is a minimum set cover problem – NP Complete We use the Quine-McCluskey optimization technique

Subset of acquaintances to query

Return


Quine-McCluskey optimization

Acquaintance

_ a

Example: A = {1,2}, B = {2,3,4}, C = {1,3} A, B, C are Prime acquaintances B is an Essential Prime acquaintance

Choose all the Essential Prime acquaintances first If any Hub is still uncovered, iteratively choose non-essential Prime

acquaintances that cover the max number of remaining Hubs, till all Hubs are covered

Return

Documents

“Sociological Orbits” Mobility Profiling and Routing for Mobile Wireless Networks