Doc.: IEEE wng Submission September 2003 Intel Corporation Slide 1 Performance of an Home Network Mesh Testbed September 15, 2003 W. Steven Conner Intel Corporation
Doc.: IEEE wng Submission September 2003 Intel Corporation Slide 2 Outline Overview of ESS Mesh Performance evaluation of a wireless home network testbed Lowering the barriers to mesh deployment Recommendation to start Mesh SG/TG Summary
Doc.: IEEE wng Submission September 2003 Intel Corporation Slide 3 Overview: Mesh Architectures Infrastructure Mode ESS with WDS Backhaul WDS Links Ad Hoc Links Peer-to-Peer Mesh (Ad Hoc Mode) Ad Hoc Links Ad Hoc or WDS Links Hybrid Infrastructure/ Ad Hoc Mesh
Doc.: IEEE wng Submission September 2003 Intel Corporation Slide 4 Overview: ESS Mesh Mesh is not limited to highly mobile networks with no infrastructure Also has application in many fixed-infrastructure environments Extended range and coverage, without requiring additional wires (convenient deployment, cost) Enhanced redundancy, reliability Potential throughput improvement Example networks where ESS Mesh is useful: Home networks, hotspot networks, etc.
Doc.: IEEE wng Submission September 2003 Intel Corporation Slide 5 Upper Level Office Lower Level Living Room Den Back Yard Question: Does it Make Sense to Deploy a Wireless ESS Mesh for a Home Network? B C D A
Doc.: IEEE wng Submission September 2003 Intel Corporation Slide 6 Overview: Experimental evaluation of an b home mesh network Upper Level Office Lower Level Living Room Den Back Yard B C D A Experiments performed in my house (~2000 sq. ft.) in Hillsboro, OR (August, 2003) Topology: 8 Client Laptops and 4 AP routers In a real home network scenario, some of the laptops would likely be replaced by other enabled devices (e.g., DVRs, media servers, stereo systems, etc.) Traffic: Experiments assume network traffic is not limited to Internet surfing on a broadband link Clients share significant amount of data within the home (e.g., A/V content sharing, photo storage, data backup, etc.)
Doc.: IEEE wng Submission September 2003 Intel Corporation Slide 7 Testbed Configurations Configuration 1 Traditional 1-hop BSS b, auto-rate, 15mW BSS emulated with ad-hoc mode All clients communicate directly with AP-A Configuration 2 Multi-hop ESS Mesh b, 11Mbps, 15mW ESS emulated with ad-hoc mode - Centrally configured minimum-airtime- metric routing (zero overhead) Clients communicate with best AP to join wireless ESS mesh Out of range
Doc.: IEEE wng Submission September 2003 Intel Corporation Slide Throughput (Mbps) OfficeLiving Room DenBackyard 70 (O) 73 (D) 75 (L) 77 (B) Multi-Hop ESS Individual Node Throughput Throughput (Mbps) OfficeLiving Room DenBackyard 70 (O) 73 (D) 75 (L) 77 (B) Individual Node Throughput Non-Mesh BSS Individual Node Throughput Out of range 1.7X 3.1X Connected!
Doc.: IEEE wng Submission September 2003 Intel Corporation Slide 9 Multi-Node Throughput Out of range Non-Mesh BSS Aggregate Throughput Multi-Hop ESS Aggregate Throughput X 1.9X 2.1X
Doc.: IEEE wng Submission September 2003 Intel Corporation Slide 10 Multi-Node Throughput cont. Aggregate Throughput with 8 Clients Out of range X
Doc.: IEEE wng Submission September 2003 Intel Corporation Slide 11 Client-to-Client Throughput Non-Mesh BSS Client-to-Client Throughput Out of range Multi-Hop ESS Client-to-Client Throughput 2.4X 3.4X Note: Direct client-to-client links can help here as well
Doc.: IEEE wng Submission September 2003 Intel Corporation Slide 12 Network Latency Non-Mesh BSS End-to-End Latency Out of range Multi-Hop ESS End-to-End Latency Highly dependent on implementation ~ 2ms increase per hop
Doc.: IEEE wng Submission September 2003 Intel Corporation Slide 13 Summary of Testbed Results A multi-hop ESS mesh is beneficial, even for a relatively small-scale home network Multi-hop topologies: Can be built with standard hardware Can improve network performance in comparison to traditional 1-hop BSS networks These experiments used 1 radio on each AP/router; multi-radio per AP/router would allow even better performance (multi-channel) Question: If mesh networking with works today, why do we need additional standards support?
Doc.: IEEE wng Submission September 2003 Intel Corporation Slide 14 Barriers to Mesh Deployment Interoperability Security Configuration / Management Should require minimal effort to deploy Lack of hooks for statistics/control Radio and metric-aware routing MAC Performance
Doc.: IEEE wng Submission September 2003 Intel Corporation Slide 15 Making Mesh Work Key areas for IEEE Standardization: Interoperability Standardizing over-the-air messaging for mesh Routing: –L2 mesh subnet for wireless backhaul –Radio and metric-aware path selection (hop-count is not sufficient!) Security: To make it possible to secure a mesh, routers should be able to trust each other Leverage/extend i for mesh Improving Configuration / Management Should require minimal effort to deploy (beyond router introduction) Statistics and control hooks need to be exposed between MAC and “mesh layer” Leverage/extend k for mesh
Doc.: IEEE wng Submission September 2003 Intel Corporation Slide 16 Research indicates MAC performance needs to be optimized for large scale mesh networks A few notable examples: RTS/CTS does not correctly solve hidden terminal problem in a mesh Tends to either sacrifice spatial reuse or allow excessive interference 1 RTS/CTS fails to achieve good schedule in a multi-hop chain RTS/CTS scheduling along a chain can cause serious TCP fairness problems and backoff inefficiencies 2 RTS/CTS does not efficiently schedule transmissions in a multi- hop chain 3 [1] Kaixin Xu, M. Gerla, and Sang Bae, "How effective is the IEEE RTS/CTS handshake in ad hoc networks?" IEEE Globecom'02, 2002, pp [2] Shugong Xu and Tarek Saadawi – “Does the IEEE MAC Protocol Work Well in Multihop Wireless Ad Hoc Networks?” IEEE Communications Magazine, June 2001, pp [3] J. Li, C. Blake, D. S. De Couto, H. I. Lee, and R. Morris. Capacity of ad hoc wireless networks. In Proceedings of ACM MOBICOM, pages , July 2001.
Doc.: IEEE wng Submission September 2003 Intel Corporation Slide 17 Enabling Mesh Usage Models Before MAC Enhancements: Home Network Small Office Small Hotspot MAC Enhancements Necessary: Enterprise Large Conference High Performance Home Network Power-users, A/V Multi-hop scheduling/scalability are significant issues
Doc.: IEEE wng Submission September 2003 Intel Corporation Slide 18 Lowering the Barriers to Mesh Deployment Standardize Multi-Hop ESS (AP Mesh) Radio/Metric-Aware L2 Routing Interoperability Security Configuration / Management Enhance MAC Performance for Mesh Scalability Scheduling (managing collisions/ interference) New Mesh Study/Task Group Leverage i/k where possible Influence current/ future MAC enhancement efforts to improve scalability for mesh Leverage e/n where possible Proposed Parallel Efforts:
Doc.: IEEE wng Submission September 2003 Intel Corporation Slide 19 Recommendation to WNG for Starting a Mesh Study/Task Group Scope: Develop an Infrastructure-Mode ESS AP Mesh that Appears as a Broadcast Ethernet to Higher Layer Protocols Scale: Up to 255 devices (APs and Clients) Security: Include support for trusted set of routers controlled by single entity Routing: Include support for both broadcast and radio/metric- aware unicast routing Multiple-radios: Include support for optional multiple-radios per router Usage Models: Initially focus on home and small- scale hotspot networks
Doc.: IEEE wng Submission September 2003 Intel Corporation Slide 20 Backup
Doc.: IEEE wng Submission September 2003 Intel Corporation Slide 21 Is IEEE the Right Place to Create a Mesh Standard? IETF/IRTF MANET groups have been working on L3 mesh standards for years But… radio awareness is out-of-scope, significantly limiting opportunity for efficient use of the wireless channel Major focus on large scale and high mobility (hard problems!) has significantly prolonged the standards process IEEE is a reasonable place to create a L2 mesh subnet standard Allows tight integration with MAC (radio awareness) Has the advantage of creating a mesh that looks like an ethernet to IP applications Improved hooks/statistics for supporting a L2 mesh can also be used to improve L3 mesh implementations IETF L3 mesh network can be used to interconnect multiple IEEE L2 mesh subnets There is recent precedent for standardizing mesh support in IEEE a already has explicit mesh support Yes, we need improved standard support for mesh in !
Doc.: IEEE wng Submission September 2003 Intel Corporation Slide 22 Fixing the MAC for Mesh We know there are issues with the current MAC, but what about e? EDCF should improve fairness and efficiency TXOPs Block ACK Direct links between clients Multiple queues allow traffic prioritization What are the implications for mesh? Improving MAC in IEEE: Option 1: Start a new study group/task group focused on MAC support for mesh Option 2: Piggyback on current/future non-mesh MAC enhancement efforts (e.g., n)