On the Design and Implementation of Infrastructure Mesh Networks - PowerPoint PPT Presentation

On the Design and Implementation of Infrastructure Mesh Networks Krishna N. Ramachandran Dept of Computer Science University of California Santa Barbara Joint work with: M. Buddhikot, G. Chandranmenon, S. Miller, E. Belding-Royer, K. Almeroth WiMesh 2005, Santa Clara

Motivation n Systems-based research on wireless mesh architectures lacking n Design and implement an architecture for infrastructure mesh networks n Offer implementation as part of NSF mesh-kit for use by other researchers Krishna Ramachandran

Outline n Motivation n MeshCluster Architecture n System Evaluation n Conclusion Krishna Ramachandran

MeshCluster Architecture AP 802.11b Internet Gateway Relay Relay 802.16 802.11a Relay 802.11a AP 802.11g 802.11a Relay AP 802.11b 802.16 Relay AP Relay AP 802.11b 802.11b Krishna Ramachandran

Design Challenges § Self-configuring, secure mesh architecture § High throughput routing § Traffic load-balancing § Seamless user mobility Krishna Ramachandran

Outline n Motivation n MeshCluster Architecture – Auto-configuration Scheme – High Throughput Routing Solution – Traffic Load-balancing – End-user mobility support n System Evaluation n Conclusion Krishna Ramachandran

MeshCluster Auto-configuration Gateway Discovery Phase: Relay runs auto-config agent 1. Relay picks zero-conf IP address 2. Agent listens to auto-config gateway advertisements 3. Advertisements contain information such as Internet 4. back-haul link speeds, DHCP server capability Relay chooses gateway using a policy (closest 5. gateway, highest back-haul capacity, etc) Krishna Ramachandran

MeshCluster Auto-configuration Relay Configuration Phase: Relay conveys information, such as number & types 1. of interfaces, observed interference, to gateway Gateway conveys back configuration parameters: 2. § Permanent IP address § ESSID for access Frequencies used on relay and access interfaces § § Power levels Relay configures itself and relinquishes temporary 3. IP address Krishna Ramachandran

Outline n Motivation n MeshCluster Architecture – Auto-configuration Scheme – High Throughput Routing Solution – Traffic Load-balancing – End-user mobility support n System Evaluation n Conclusion Krishna Ramachandran

MeshCluster Routing Design Choices: n Layer-2 or Layer-3 routing? n How about Layer-3 wireline protocols? n Existing ad hoc routing protocols? Krishna Ramachandran

Why not AODV (or any other reactive MANET protocol)? n Lack of routing metric support n Route maintenance mechanism missing n Route discovery latency even for common- case traffic n Routing table size can be very large in wireless mesh deployment Krishna Ramachandran

AODV-ST extends AODV n Supports spanning trees rooted at gateways. Several advantages: – Eliminates route discovery latency for common- case traffic – Routes to gateways are automatically maintained – Relays lie on best paths to gateway according to metric, such as ETX, ETT, etc. n Tunnels out-bound packets to gateway without route discovery as with AODV Krishna Ramachandran

AODV-ST Spanning Trees Gateway 2 Gateway 1 ST-1 Link ST-2 Link R R R R R R R Krishna Ramachandran

Outline n Motivation n MeshCluster Architecture – Auto-configuration Scheme – High Throughput Routing Solution – Traffic Load-balancing – End-user mobility support n System Evaluation n Conclusion Krishna Ramachandran

Traffic Load-balancing Defined Two types: n Path load balancing Source Destination – Route coupling can limit performance gains n Gateway load balancing – Route coupling Gateway Source Gateway less of an issue Krishna Ramachandran

Gateway Load-balancing n Access relay lies on multiple STs corresponding to multiple gateways n Chooses best gateway as default gateway n Sends RTT probes periodically to all gateways n If (least-delay gateway != default gateway) – Possible congestion on path to default gateway – So choose least-delay gateway as egress point n Currently investigating route flapping issues Krishna Ramachandran

Outline n Motivation n MeshCluster Architecture – Auto-configuration Scheme – High Throughput Routing Solution – Traffic Load-balancing – End-user mobility support n System Evaluation n Conclusion Krishna Ramachandran

End-user Mobility Support n Mobile IP based domain mobility n Simple DHCP based mobility n MobileNAT based mobility Krishna Ramachandran

Mobile IP based mobility Home Agent Gateway Internet Relay Relay 802.11a FA AP 802.11a Relay 802.16 Relay FA AP Relay FA AP 802.11b 802.11b Krishna Ramachandran

Simple DHCP Mobility Support Mobility Mgr Gateway Internet Relay Relay Mobility 802.11a Agent AP 802.11a Relay 802.16 Relay Mobility Agent AP Relay Mobility agent AP 802.11b 802.11b Krishna Ramachandran

Outline n Motivation n MeshCluster Architecture – Auto-configuration Scheme – High Throughput Routing Solution – Traffic Load-balancing – End-user mobility support n System Evaluation n Conclusion Krishna Ramachandran

Prototype Implementation Specifications: 100/133 Mhz AMD ElanSC520 on Soekris net4521 board n Mini-PCI type III socket n 2 PC-Card/Cardbus slots, for wireless adapters n Debian Linux 2.4.26 n AODV-ST routing protocol n Modified hostap Prism2 driver n Auto-config tools, DHCP, NAT, Firewall, RADIUS, Web server, n MobileIP, Simple IP Krishna Ramachandran

System Evaluation Goals Evaluate individual components of the architecture: n Mesh Auto-configuration n AODV-ST Routing performance n Load balancing solution n Mobility schemes Krishna Ramachandran

Mesh Auto-configuration Simulation Setup n 30 mesh relays randomly distributed in a terrain of 1000x1000m n Single-radio IEEE 802.11a mesh relays n Relays randomly pick a channel upon startup n One gateway in the network issues periodic gateway advertisements n Implemented layer 2 BSSID/ESSID beaconing n 0-40 interfering devices in increments of 10 n Interfering devices randomly pick channel and ESSID to operate on Krishna Ramachandran

Mesh Auto-configuration Results Krishna Ramachandran

AODV-ST Evaluation Methodology UCSB MeshNet MeshNet Devices Krishna Ramachandran

AODV-ST Results Krishna Ramachandran

Mobility Scheme Evaluation lasthop1 lasthop Internet G A B C Channel – 11 Pseudo-IBSS mode Transmit Rate – auto (ARF) Krishna Ramachandran

Mobile IP Evaluation Mobility associated delays Association Delays Router Solicitation MIP Reg Response MIP Reg Request AP Association Start 0 100 200 300 400 500 600 700 800 Time (milliseconds) Krishna Ramachandran

Simple IP Evaluation Mobility associated delays Association Delays AP Association DHCP Discover DHCP Request DHCP Offer DHCP ACK Start 0 1000 2000 3000 4000 5000 6000 Time (milliseconds) Krishna Ramachandran

Conclusion & Future Work MeshCluster addresses four key design challenges: n – Mesh auto-configuration – High-throughput routing – Traffic load-balancing – User mobility support Future work: n – Multiple radio support to improve capacity – Routing protocol for multi-radio wireless mesh networks – Integration of monitoring/management utilities into MeshCluster architecture – NSF MeshKit Krishna Ramachandran

Krishna N. Ramachandran krishna@cs.ucsb.edu AODV-ST download link: http://www.cs.ucsb.edu/~krishna/aodv-st/ Krishna Ramachandran

Mesh Network Types Client-Mesh Infrastructure-Mesh Networks Networks End-devices End-devices do not participate in participate in forwarding forwarding Typically lack a Service providers service provider manage the network managing the network Krishna Ramachandran

MobileNAT Mobility Support Krishna Ramachandran