PeopleNet: Engineering A Wireless Virtual Social Network
Authors: Mehul Motani, Vikram Srinivasan and Pavan S. Nuggehalli
Presented by: Sharon Smith
Motivation:Observation: People often use social contacts for: time, location and community-specific information rather than using powerful search engines or libraries! Social contacts are generally good sources of this information.
Inference: There is a need for a simple and efficient mechanism to find location, time, and community-specific information between people.
PeopleNet Solution:
Solution: A seamless wireless virtual social network utilizing an architecture called PeopleNet to meet the needs of these information seekers.
PeopleNet is a query matching system exploits the:
(1) Natural behaviors of social networking and social mobility
(2) Pervasiveness of mobile phones and their p2p capabilities
PeopleNet System Perspective:
PeopleNet has 4 KEY advantages:
Convenient
$$ Cost-savings
Scalable
Lightweight
PeopleNet System Description: Main components:
Long distance fixed infrastructure Short distance p2p interfaces
Cellular or WiMaxBluetooth or WiFi
Divide area into non-overlapping regions called bazaars
Each bazaar handles specific types of queries
PeopleNet System Description:
Bazaar 1
Bazaar 2
Bazaar 3
Bazaar 4
Bazaar 5
Bazaar 6
Fixed cellular infrastructure
PeopleNet clustered cells form 6 bazaars
PeopleNet Fixed Infrastructure:
Reference: www.althos.com
A bazaar would consists ofseveral base stations controlledby a MSC = Mobile Switching Center
A PC = PeopleNet Coordinatorwould be located at the MSCto provide features for PeopleNet
Incoming Query:PC chooses k users from MSC’s VLR= Visitor Loation register and transmits query
Outgoing Query:PC with LUT maps query type to respective bazaar’s PC
Query Hierarchical Format:
Query description is hierarchical with i higher than j, i > j
A match occurs when all the specified levels of the requestare identical to the response
Propagating Queries in P2P Mode:Random Spread:
nA nB
1(A)3(Q)3(Q)…
1
2 3(Q)
5(Q)
…
3(Q)
4nA’s Buffer nB’s Buffer
5 5(Q)
Random Swap:
nA nB
1(A)3(Q)
3(Q)… 1 2 3(Q)
5(Q)
5(Q)…
3(Q)3
nA’s Buffer nB’s Buffer 4 5(Q)
5(Q) 6
3(Q)3
3(Q)5
5(Q)
6
PeopleNet’s Metrics for Analysis:1) Probability of match
2) Time to match
3) Time in system
4) Number in system
5) Number of distinct matches
QoS for user
Useful in buyer/seller application
Rate to re-inject queries
PeopleNet’s Review Notation: Guide to PeopleNet’s notation and metrics:
N= Number of nodes in a bazaar= Arrival rate of new queries in the system
2M= Total number of query typesk= Total number of nodes in a bazaar that receive query
from cellular networkB= Size of buffer at each nodeL= Number of queries exchanged when nodes meetW*W= Square grid distance of location, W*W/m= bazaar size
PeopleNet’s Review Assumptions:
1) Every node occupies one grid position
2) Nodes can move either: random walk or i.i.d. walk
3) For every type of query this is a unique matching query type
4) Time is discrete
5) Probability of queries arriving =
6) Query types are U (1, 2M)
7) Arrival rates are randomly distributed to k with N nodes
8) Transmission radius is sqrt(2) units
Swap vs. Spread Comparison:Metrics utilized for RePast simulations:
W=32, N=30, B=3, M=30, lambda=.5
Time in System Avg. Copies
Swap vs. Spread Comparison:Metrics utilized for RePast simulations:
W=32, N=30, B=3, M=30, lambda=.5
Matching Probability No. of Distinct Matches
Qualitative Impact of Bazaars:Theorem 1: The expected number of queries of a certain type will be
constant given by:
The mean is independent of k and lambda, however the variance isnot.
Smaller bazaar performs better!
Unfortunately, due to mobility patterns in real life, bazaars can’t be made very small.
Swap vs. Spread Analysis:
Where,
Swap Probability of Match:Probability of Match vs. Buffer Size
As Buffer Size increases, matching prob.
increases. Its maximal values occurs when L = B/2.
Recall, L= number queries swapped when two
nodes meet!
Meta-information Exchange:Random Swap:
nA nB
Y(Q)
X(Q) How do we decide which and how many queries to exchange?
Maybe nodes could “employ some intelligence in swapping queries by
prioritizing.”
Meta-information Exchange:
Using a summary snapshot of each node’s data, known as meta-information the average number of distinct queries can beimproved.
Given buffer information:
Bx= [3Q, 3Q, 3Q, 4A]By= [3A, 3A, 4Q, 4Q]
The meta-information would look as follows:
X= [(3Q, 3), (4A, 1)]Y= [(3A, 2), (4Q, 2)]
Meta-information Exchange:
Greedy meta-rank:
Only takes into account highest count and chooses tomatch this way.
Problem: Possibility of exchanging queries that it already matched.
Smart meta-rank:
Uses a thought experiment, by choosing largest query typefirst and then reorders the meta-information of peer and
finds new largest query type count.
Greedy vs. Smart Meta-rank Algorithms:Greedy:
Smart:
10 matches!
11 matches!
Naive vs. Smart Results for Swap:
Method Matching Probability Time to MatchNumber of
Matches
Naïve Random Swap 0.73 250 0.25
Smart Meta-rank 0.85 150 2
With the following parameters:N=30, k=5, W=32, M=500, B=50, lambda=.5
12% better 40% reduced 8x
Drawbacks and Extensions:System Security:
Data IntegrityViruses and worms targeted to mobile phones
Buffer Management (deleting queries in buffer):Time to Date (TTD) Matches to Date (MTD)
System Capacity:Depends on lambda and M parametersClear understanding necessary for maximal benefits
Swap vs. Spread:Infinite Buffers vs. Finite Buffers
Query Database:Need to decide query types and inform users
Drawbacks and Extensions Cont’d:Mobility Patterns
Random walk mobility model is not ideal for mobile usersNeed to find more accurate mobility model
Bluetooth or P2P pervasivenessIs it fair to assume all mobile phones have p2p capability?
Persuading users to support the PeopleNet architectureWhat benefits does a relaying mobile phone user have?Why should they bother?
Questions/Comments?