06 Polychronis Yapdimatopoulos -- igbos g_

Social Networking and file sharing in downtown Boston

Polychronis [email protected]

Polychronis Ypodimatopoulos <[email protected]>MIT Media Lab

How do we meet/find people

● ... when wandering downtown?● ... in my neighborhood?● ... in my office?

mailto:[email protected]






Why not?

● Data transfer must be efficient

(forget about TCP/IP)

● Discovery must be scalable

(humans/objects/applications)

● Overhead and Power Consumption must be minimized



Data transfer: 1958 2008

http://www03.ibm.com/ibm/history/exhibits/specialprod2/specialprod2_5.html



Scalable discovery



Cerebro: http://cerebro.mit.eduProvides:

● presence information (including distance) for all other users/objects in the network

● mesh networking on regular 802.11b/g devices

● extensible user profile including nickname, colors, keys, IP addresses, pictures, status message etc

● file sharing using an efficient multicast mechanism

● simple collaboration mechanism

● connect to remote mesh networks over the internet

● interoperability among different architectures (x86, ARM)

● programming API



Cerebro: discovery



Cerebro: discovery



Cerebro: discovery



Profile arrival times



CDF of profile arrivals



Protocol overhead – Convergence time



Cerebro: File sharing

How “illegal” can we get?



Cerebro: File sharing

TCP/IP: sequential Cerebro: broadcast(opportunistic and reliable)



Transfer of 2MB to 27 nodes

Cerebro: 1Mbps broadcast,

simultaneous

TCP/IP: 10Mbps, unicast, sequential



Transfer of 2MB to 27 nodes



Platforms tested

● x86 ● ARM (Nokia N800, Gumstix)● OpenMoko (Freerunner)

Requirements:● Python● Dbus



Cerebro GUI (x86, OpenMoko)Who What



http://cerebro.mit.edu



Why multiple entries? Mobility!

Node ID Witness Distance SourceC C 1 CD C 3 CE D 4 CF D 4 CG D 4 CD D 2.5 DE D 3.5 DF D 3.5 DG D 3.5 DE E 5 E

Presence information arriving at B:



Presence Table

Node ID Witness Distance SourceB B 1 BC C 2 BD C 3 BE C 4 BF C 4 BG C 4 B

Presence information arriving at A:


May 2008 Polychronis Ypodimatopoulos <[email protected]>MIT Media Lab

24

1) Wait for a period T for presence beacons from neighboring nodes2) For every beacon received:

For every entry in beacon:• if it is about the current node, or the current node is the witness, or the the serial number is not newer than the existing one, discard it!• if node/witness pair exists in presence table update number of arrivals and next arrival estimate both for node and witness• else add new entry in the table

4) Eliminate stale entries in presence table (ie. entries where the next arrival estimate has lapsed)

5) Create a new beacon using the minimum distances to each node in presence table

6) Broadcast beacon to neighbors

Presence Protocol



25

Node ID Witness Distance Serial No.

ID of the node this entry is about

(MAC address, 6 bytes)

ID of the node from which this entry was

received (MAC address, 6 bytes)

Unique ID for each entry published by

node <Node ID>(2 byte)

Distance from the current node to <Node ID>

(similar to ETX metric)(1 byte)

Presence Protocol for Large Mesh Networks

Presence Frame contents



26

Distance calculation

Distance D:D = T/AT: period of timeA: number of arrivals



27

Arrival estimates

1)Count arrivals over time period T.2)Formulate a Poisson arrival rate (assuming arrivals are

independent events)3)Estimate time of next arrival with 90% accuracy

Example: For T=1sec, accuracy 90%, next arrival must be in 2.3sec

For T=1sec, accuracy 100%, next arrival must be in infinity (!)


06 Polychronis Yapdimatopoulos -- igbos g_

Technology

transferof2mbto27nodes

mbps broadcast

broadcast tcpip

simultaneous tcpip

ip addresses

filesharing howillegalcanweget