How Small Labels create Big Improvements April. 2013 Chan-Myung Kim LINK@KoreaTech .

How Small Labels create Big Improvements

April. 2013

Chan-Myung KimLINK@KoreaTech

http://link.koreatech.ac.kr

ABSTRACT Identifying communities in an ad hoc mobile communica-

tions system, such as a PSN, can reduce the amount of traf-fic created when forwarding messages.

But there has not been any empirical evidence available to support this assumption to date.

In this paper, we show through use of real experimental human mobility data, how using a small label, identifying users according to their affiliation, can bring a large im-provement in forwarding performance, in term of both de-livery ratio and cost.

LINK@KoreaTech 2

INTRODUCTION Pocket Switched Networks (PSN) represent one particular

intermittent communication paradigm for mobile radio de-vices.

In the research community, it has been a widely held belief that identifying community information about recipients can help select suitable forwarders, and reduce the delivery cost compared to naive “oblivious” flooding.

However, to date as far as we are aware, there has been no experimental evaluation of this belief, and no-one knows whether it is valid or not.

LINK@KoreaTech 3

INTRODUCTION We created a human mobility experiment during IEEE Info-

com 2006, with the participants labelled according to their academic affiliations.

After collecting 4 days of data during the conference period, we replayed traces using an emulator, and we discovered that a small label can indeed effectively reduce the delivery cost, without trading off much against delivery ratio.

The intuition that simply identifying community can im-prove message delivery turns out to be true even during a conference where the people from different sub-communi-ties tend to mix together.

LINK@KoreaTech 4

EXPERIMENTAL SETUP The device used to collect the contact opportunity data and

mobility statistics in this experiment is the Intel iMote(ARM processor, Bluetooth radio and flash RAM).

We packaged these devices in a dental floss box, due to its ideal size, low weight, and hard plastic shell.

Eighty of these boxes were distributed to attendees at the IEEE Infocom conference in Barcelona in April 2006.

The participants are specially selected so that thirty-four out of eighty form four subgroups according to academic af-filiations.

Paris(4,10), Switzerland(5), Barcelona(15)

LINK@KoreaTech 5

EXPERIMENTAL SETUP The iMotes were configured to perform a Bluetooth base-

band layer inquiry discovering the MAC addresses of other Bluetooth nodes in range, with the inquiry mode enabled for five seconds.

Between inquiry periods, the iMotes were placed in a sleep mode in which they respond to enquiries but are not other-wise active, for a duration of 120 seconds plus or minus twelve seconds in a uniform random distribution.

The randomness was added to the sleep interval in order to avoid a situation were iMotes timers were in sync, since two iMotes performing inquiry simultaneously cannot see each other.

LINK@KoreaTech 6

ANALYSIS INTER-CONTACT TIMES Inter-contact times distribution is a good indication for rela-

tionship.

Inter-contact time follows a power-law distribution, the big-ger that value of the power coefficient, the more frequently the nodes pair interact.

We extend to look at inter-contact distribution for all the nodes inside a group and also the inter-contact distribution between two groups.

We believe the power-law coefficient of these inter-group in-ter-contact time distribution, if they are following power law, indicate the closeness of two groups.

LINK@KoreaTech 7

ANALYSIS INTER-CONTACT TIMES .

LINK@KoreaTech 8

ANALYSIS INTER-CONTACT TIMES .

LINK@KoreaTech 9

ANALYSIS INTER-CONTACT TIMES Here we also want to introduce the concept of friendship

communities(not same group but same building).

People from one group may be good forwarders for people in the corresponding friendship group.

In our experimental data, we happen to have two groups from Paris, and we want to look at whether they have closer relationship when compared to other groups, based on the inter-contact time distribution.

LINK@KoreaTech 10

ANALYSIS INTER-CONTACT TIMES .

LINK@KoreaTech 11

EVALUATION METHODOLOGY In order to evaluate algorithms, we developed an emulator

called HaggleSim, which can replay the mobility traces and emulate different strategies on every contact event.

In all the simulations in this work, we divided the traces into discrete contact events with an granularity of 100 seconds.

LINK@KoreaTech 12

Simulation Parameters Number of copies: The maximum number of duplicates of

each message created at each node.

Number of hops: The maximum number of hops, counted from the source, that a message copy can travel before reaching the destination; this is similar to TTL value in the Internet.

Time TTL: The maximum time a message can stay in the system after its creation. This is to prevent expired mes-sages from further circulation.

LINK@KoreaTech 13

Performance Metrics Delivery ratio.

Half-life delivery time TTL: the time TTL value that would al-low half of the messages created to be delivered; It mea-sures how fast and efficient a forwarding strategy for mes-sages delivery.

Hop-distribution for deliveries: the distribution of the num-ber of hops needed for all the deliveries

Delivery cost: total number of messages (include dupli-cates) transmitted across the air.

LINK@KoreaTech 14

Simulation Scenario We created the following scenario: all the 77 nodes create

1000 messages, destined only to the 34 nodes belonging to the four groups; the message creation times are uniformly distributed throughout the experimental duration.

To ensure that the performance improvement is not due to randomly limited number of forwarders, for every round of simulation, we created four random groups of same group sizes as the original groups but with nodes randomly se-lected from all the 77 nodes.

LINK@KoreaTech 15

RESULTS AND ANALYSIS .

LINK@KoreaTech 16

RESULTS AND ANALYSIS .

LINK@KoreaTech 17

RESULTS AND ANALYSIS .

LINK@KoreaTech 18

RESULTS AND ANALYSIS .

LINK@KoreaTech 19

RESULTS AND ANALYSIS .

LINK@KoreaTech 20

RESULTS AND ANALYSIS .

LINK@KoreaTech 21

CONCLUSIONS Just an affiliation label is shown to bring significant im-

provement in forwarding performance over oblivious or naive forwarding algorithms in PSNs.

We need to embed this kind of state information in our fu-ture designs for forwarding strategies, and we have shown that labels that identify community provide a good start.

LINK@KoreaTech 22