Gateway Forwarding Strategies in Ad hoc Networks ADHOC’04, Johannesberg Erik Nordstr¨ om, Per Gunningberg IT department, Uppsala University Christian Tschudin CS Department, University of Basel
The story so far. . . Ad hoc ≈ Mobile Multi-hop Wireless Networking IETF started MANET (Mobile Ad hoc NETworks) working group on ad hoc routing in 1995/96. Since then: • Four routing protocols going for RFC status ( AODV , OLSR , DSR , TBRPF) • AODV implementations ≈ 10 • Zero deployment One explanation: No Internet integration � Erik Nordstr¨ c om, Uppsala University ADHOC’04, Johannesberg 2/16
Common MANET Internet connectivity scenarios • Extending the reach of wireless LANs • Disaster area rescue drones with connection to rescue central • Military units communicating with headquarters – Uppsala students are building rescue robots with ad hoc routing for participation in Robocup. � Erik Nordstr¨ c om, Uppsala University ADHOC’04, Johannesberg 3/16
So what has gone wrong? Internet and ad hoc networks are very different: • Hierarchical vs. Flat addressing and routing • Fixed vs. Mobile (macro and micro mobility) For Internet connectivity many suggest Mobile IP (MIP). • It is only half the solution (solves macro mobility) • Provides topologically correct (care-of) addresses • Operates (mainly) at gateways towards the Internet (i.e., MIP does not really matter until packets reach the gateway) � Erik Nordstr¨ c om, Uppsala University ADHOC’04, Johannesberg 4/16
Example of an Internet connected ad hoc network Combining Ad hoc Networking with MIP connectivity: 66.35.250.151 • Computers are mobile and use their home 130.238.12.1 addresses HA Internet 130.238.8.1 192.168.6.1 MIP Tunnel FA 192.168.1.1 • Multiple gateways (MIP agents, FA, HA) to 10.0.4.2 142.67.8.245 the Internet 65.43.32.1 MN 130.238.12.5 63.3.5.23 • Multi-hop paths to gateways How do we stitch all this together? � Erik Nordstr¨ c om, Uppsala University ADHOC’04, Johannesberg 5/16
Challenges • Gateways have to be discovered and self-configured at nodes • Gateway changes have to be trackable so that MIP can re-register (otherwise return traffic is lost) • For smooth operation, hand-over and multi-homing might be beneficial (i.e., we must support connectivity to more than one gateway at once) • Efficient integration with the routing protocol is important (for example, mixing proactive gateway discovery with reactive routing does not make sense) � Erik Nordstr¨ c om, Uppsala University ADHOC’04, Johannesberg 6/16
Our contribution A lot of Previous Work on: • Address configuration • Gateway discovery (efficiency in gateway advertisements) Our complementary work is a comparison of different gateway forwarding strategies: • Default routes – one generic path to a gateway for all external (Internet) destinations. • Tunneling – temporary replace the destination address with the gateway address � Erik Nordstr¨ c om, Uppsala University ADHOC’04, Johannesberg 7/16
Default Routes – Routing Tables Destination Next Hop Hop Cnt Destination Next Hop Hop Cnt 63.3.5.23 63.3.5.23 1 192.168.1.1 63.3.5.23 3 _ 66.35.250.151 default 63.3.5.23 63.3.5.23 1 _ default 63.3.5.23 3 66.35.250.151 default default 192.168.1.1 3 (a) (b) a) we need host state (to avoid subsequent route lookups on source and intermediate nodes). . . b) but also a gateway route to track gateways (Globalv6 proposal) • as much as three routing table lookups might be necessary (host → default → gateway/next hop) � Erik Nordstr¨ c om, Uppsala University ADHOC’04, Johannesberg 8/16
Default Routes – State Replication Problem Extra routing table state (host, gateway, default route) introduces problems: • states (e.g., S A ) must be replicated on intermediate nodes • how do we handle repairs or route optimizations? D D S A S A S A S A S A A B C GW A B C GW (b) (a) � Erik Nordstr¨ c om, Uppsala University ADHOC’04, Johannesberg 9/16
Default Routes – Gateway Tracking Problem A default route might not stay consistent along a hop by hop path when there are multiple gateways • Breaks two-way communication (e.g., TCP) RREQ RREP GW1 GW1 A A GW1 A C C C mismatch B B B GW2 GW2 GW2 Default route pointing to GW1 Default route pointing to GW2 � Erik Nordstr¨ c om, Uppsala University ADHOC’04, Johannesberg 10/16
Tunneling – Routing Tables Gateway 192.168.1.1 130.238.8.1 Decapsulation Internet Encapsulation 63.3.5.23 63.42.32.1 Ad hoc src node 130.238.12.5 Tunnel Destination Next Hop Hop Count Flags Destination Next Hop Hop Count Flags 66.35.250.151 66.35.250.151 192.168.1.1 1 I 192.168.4.3 65.3.5.23 2 63.3.5.23 63.3.5.23 1 63.3.5.23 63.3.5.23 1 192.168.1.1 63.3.5.23 3 G 192.168.1.1 192.168.1.1 1 Destination Next Hop Hop Count Flags Destination Next Hop Hop Count Flags 192.168.4.3 192.168.4.3 1 192.168.4.3 65.43.32.1 3 65.43.32.1 65.43.32.1 1 65.43.32.1 65.43.32.1 1 192.168.1.1 65.43.32.1 2 • Gateways addressed in packets – not in nodes (some overhead) • No extra state at intermediate nodes – only a gateway route • Can be done one-way to gateway only � Erik Nordstr¨ c om, Uppsala University ADHOC’04, Johannesberg 11/16
Tunneling – Benefits • Protocol transparency • Route aggregation • Stability (no gateway tracking problem) • Multiple gateways (multi-homing, load-balancing, hand-over) (a) (b) (c) Figure 1: (a) Default route. (b) & (c) Tunnel configurations. � Erik Nordstr¨ c om, Uppsala University ADHOC’04, Johannesberg 12/16
Comparison ns-2 simulations with: • Two gateways, 10 - 20 mobile nodes, fixed density • Tunneling to gateways • Globalv6-draft style default routes • Modified default routes – A reference implementation that forwards all traffic on a default route, ignoring host route state (we have no traffic among ad hoc nodes) � Erik Nordstr¨ c om, Uppsala University ADHOC’04, Johannesberg 13/16
Some simulation results – CBR Traffic (UDP) 100 12 Tunneling Default routes Default routes mod. 95 10 Control traffic / data (%) CBR Delivery ratio (%) 90 8 85 6 80 4 75 Tunneling 2 Default routes Default routes mod. 70 0 8 10 12 14 16 18 20 8 10 12 14 16 18 20 Number of nodes Number of mobile nodes • Default routes have state replication problems – ignoring state helps � Erik Nordstr¨ c om, Uppsala University ADHOC’04, Johannesberg 14/16
Some simulation results – TCP Traffic (FTP) 0.4 10 Tunneling 9 Default routes 8 TCP Throughput (Mbps) Control traffic / data (%) Default routes mod. 0.3 7 6 0.2 5 4 3 0.1 2 Tunneling Default routes 1 Default routes mod. 0 0 8 10 12 14 16 18 20 8 10 12 14 16 18 20 Number of nodes Number of mobile nodes • With TCP we also need to track gateways – ignoring state does not help as much here � Erik Nordstr¨ c om, Uppsala University ADHOC’04, Johannesberg 15/16
Conclusions Internet connectivity is critical for MANET deployment. Our comparison shows: • The default route concept does not transfer well to MANETs • Default routes operate incorrectly with multiple gateways • Tunneling is efficient, simple, transparent and works with multiple gateways at the cost of a small overhead We will demo ad hoc network Internet connectivity today! � Erik Nordstr¨ c om, Uppsala University ADHOC’04, Johannesberg 16/16
Recommend
More recommend
