6.964 Pervasive Computing Grid: Scalable Ad Hoc Networking 1 November 2001 Douglas S. J. De Couto Parallel and Distributed Operating Systems Group MIT Laboratory For Computer Science

Who are we? Grid project in PDOS Professor: Robert Morris Students: –Douglas De Couto –Dan Aguayo –Jinyang Li –Ben Chambers –Hu Imm Lee

Outline Motivation “Classic” ad hoc protocol Geographic forwarding Grid location service (GLS) Location proxies The Grid network

So you want to build a pervasive network? Assumptions –Wireless, packet-based, mobile –Bigger than just your living room (multihop) Today’s approach: IEEE base stations –Site survey, measure radio performance –Channel Allocation –Inter-base-station network (wiring?) –…

B1 Base-station example B2 4 Wired network 2 1 3

Ad hoc: a better way “Ad hoc” means no infrastructure, no planning –Normally implies wireless, mobile, multihop Place devices (nodes) anywhere Constraint: devices should form connected network –If not, add “relay nodes” Costs less!

Ad hoc example r

Ad hoc scenarios Temporary, fast setup –Emergencies –Social events Rooftop networks –Connect neighborhoods –No wires, trenches, etc. Developing communities –Ad hoc is cheaper, more incremental –Automatic protocols  no technicians needed

Other ad hoc benefits Better spectrum reuse (spatial) Better scalability Possibly better power

Ad hoc challenges How do we find multihop routes? Is there enough network capacity? Does it use too much device power? –Span: Chen et al., Mobicom 2001

“Classic” protocol Dynamic Source Routing (DSR) –Flooding route discovery finds source routes as needed –Aggressive caching helps performance

Why not use DSR? Protocol works well with about a hundred nodes –Simulation results; vary with movement, data traffic Protocols scales poorly –Propagates topology information throughout network –Overhead grows too fast with network size, especially with mobility

DSR overhead Number of nodes Avg. packets transmitted per node per second

Geographic forwarding (GF) Packets addressed to  id,location  Next hop is chosen from neighbors to move packet geographically closer to destination location Routing overhead constant as network size (nodes, area) grows A B C D F C’s radio range E G

A B C D E F A’s nbrs: B, 1 hop (nh: B) C, 2 hops (nh: B) B’s nbrs: A, 1 hop (nh: A) C, 1 hop (nh: C) D, 2 hops (nh: C) Local protocol is 2-hop distance vector A sends packets to F d cf > d bf but… d df < d bf and C is B’s next hop to D D, 2 hops (nh: C) GF With a Local Protocol

Geo. forwarding challenges How do we find destination locations? How do nodes find their own locations? –Location sensors not always practical Topology problems (“holes”) General ad hoc problems –Power, capacity

A E H G B D F C J I K L Each node has a few servers that know its location. 1. Node D sends location updates to its servers (B, H, K). 2. Node J sends a query for D to one of D’s close servers. “D?” Grid Location Service (GLS) overview

Grid Node Identifiers Each Grid node has a unique identifier. –Identifiers are numbers. –Perhaps a hash of the node’s IP address. Identifier X is the “successor” of Y if X is the smallest identifier greater than Y.

level-0 level-1 level-2 level-3 All nodes agree on the global origin of the grid hierarchy GLS’s Spatial Hierarchy

3 servers per node per level n s s s ss s s s s s is n’s successor in that square. (Successor is the node with “least ID greater than” n ) sibling level-0 squares sibling level-1 squares sibling level-2 squares

Queries search for destination’s successors Each query step: visit n’s successor at increasing level. n s s s s s s s s s3 x s2 s1 location query path

Geographic forwarding is less fragile than source routing. DSR queries use too much b/w with > 300 nodes. Fraction of data packets delivered successfully Number of nodes DSR Grid GF + GLS performs well Biggest network simulated: 600 nodes, 2900x2900m (4-level grid hierarchy)

GLS properties Spreads load evenly over all nodes Degrades gracefully as nodes fail Queries for nearby nodes stay local Per-node storage and communication costs grow slowly as the network size grows : O(log n), n nodes More details: Li et al, Mobicom 2000

Geo. forwarding challenges How do we find destination locations? How do nodes find their own locations? –Location sensors not always practical

Location Proxies Nodes that know their location can act as location proxies Location proxies can communicate with each other using geographic forwarding and the local routing protocol Nodes without location select proxies, and communicate through them using the local protocol Nodes advertise proxy locations as their own Proxies not special besides knowing locations

Proxies Increase Delivery Rate

The Grid network Red: 5 th floor Blue: 6 th floor > 20 “relay” nodes About 2 to 4 hops across each floor

Current Grid services IP routing, including Internet gateway –E.g. supports traceroute Grid specific information –Who can my radio talk to? –Who do I have routes to?

Grid services “in progress” Location service –Where is node X? Geocast –Send message m to every node in region R Position estimation protocol –I don’t have a position sensor –Where am I?

Grid Applications What is a “Grid application” –Uses unique Grid services Under development: Grid chat –Regular text + voice chat –Who’s nearby? (ask Grid) –Who’s at the student center? (ask Grid)

Grid details Protocol software implemented in the Click modular router –Runs at userlevel, easy to interface to applications –Very portable Nodes: –Mobile: iPaqs PCMCIA + Linux –“Relay” : small PCs PCI cards + OpenBSD Global distance vector (DV), or k-hop DV + GF

Grid Summary Grid routing protocols are: –Self-configuring –Easy to deploy –Scalable ipkg: ipkg install grid