Carnet: Scalable Ad-Hoc Mobile Networking Robert Morris with Kaashoek and Karger
How can a network support a world of devices? No wires. Mobility. Self-configuration. Ease of deployment. Scalability to 1000s of nodes.
Ad-Hoc Nets: The Dream Nodes forward each others’ packets. No infrastructure; easy to deploy; fault tolerant. Short hops are good for power and spectrum. Can it be made to work? Hari Frans
Ad-Hoc Nets: The Reality From a simulation of the best existing ad-hoc system. Y axis is protocol overhead, in packets per second. Node motion causes protocol activity.
Carnet Overview Radio Neighbors Geographic Forwarding Location Server Location Client AppsIP
Geographic Forwarding Scales Well A B C D F C’s radio range A addresses packets to G’s latitude/longitude. G is “east” of C. C only needs to know about immediate neighbors. E G
Grid Location Service (GLS) A G … G … G f(G) G updates all of f(G) as it moves Query to f(G) G registers its latitude/longitude with the set of nodes f(G). Any other node can compute the same f(G) to send queries. Self-contained, load-balanced, fault-tolerant. X Y Z
GLS’s Fixed Spatial Hierarchy Level 1 Level 2 Level 3 Level 4
3 Servers Per Node per Level n s s s s s s s s s In each square, n’s server is the node with ID numerically closest to n.
Queries Search Upwards n s s s s s s s s s3 Each step: visit node with ID closest to n in surrounding square. x s2 s1
The Induction Assumption “Route to node with ID closest to x” works in each Level L square. 3 Level L Square Update from 18 Level L+1 Square
The Inductive Step (1) Each node sends an update to its server in sibling squares at Level L. 3
The Inductive Step (2) Result: each node knows all nodes with close IDs at Level L
Route to Node with ID near X= Update from 18
Simulation Environment A square universe. 100 nodes per square km. No locality: –Nodes move randomly at 0 to 10 m/s. –Uniform communication.
Carnet Overhead Scales Well
Carnet Delivers Packets
Challenges Variable node density. Communication patterns. Per-node available b/w. Geographic holes. Spatial congestion control.
Carnet Summary Self-configuring. Easy to deploy. Scalable. Starting to build prototype.