200 likes | 324 Views
Wireless “ESP”: Using Sensors to Develop Better Network Protocols. Hari Balakrishnan Lenin Ravindranath , Calvin Newpor t, Sam Madden M.I.T. CSAIL. Big Changes in Access Devices. Smartphones will generate half of mobile data traffic this year
E N D
Wireless “ESP”: Using Sensors to Develop Better Network Protocols Hari BalakrishnanLenin Ravindranath, Calvin Newport, Sam Madden M.I.T. CSAIL
Big Changes in Access Devices • Smartphones will generate half of mobile data traffic this year • Smartphones and tablets will exceed PC sales by 2011 • 172M smartphones sold worldwide in 2009 • 25% of US phone market; 50% in two years
The Problem • Users with “truly mobile” devices • Switch between static and mobile modes (and move across different environments) • Most protocols optimized for static settings • Protocols that compensate for mobility are not optimal for static settings Need protocols to adapt to both settings over short periods of time
Static v. Mobile • Channel relatively stable • Protocols can average estimates • Protocols should ignore short-term variations • Network topology is unchanging • Protocols can probe less frequently • Channel changes fast • Channel assessments quickly outdated • Protocols should not maintain long histories • Network topology changes more rapidly • Probe more often • Optimal protocols are different for static and mobile settings
Example: Different Loss Patterns Probability that packet i+k is lost given packet iis lost 10 ms k
Pop Quiz AP Client 1 Client 2 Client 2 leaves range of AP at around t=35 seconds Client 1’s throughput drops for several seconds Why?
Today’s Protocols • Attempt to adaptimplicitly using measurements of packet loss, BER, SNR • Difficult adaptation problem for truly mobile devices • Lack explicit knowledge about the prevalent mobility mode • Can we do better?
The Opportunity Accelerometer Proximity Sensor Camera • Modern mobile devices have many sensors • Used by applications • Ignored by protocols today WiFi Bluetooth Wireless protocols can use hints from sensors to significantly improve performance GPS Ambient Light Sensor Microphone
Sensor Movement Hints • Has there been movement? • Heading (direction) • Speed • Position 50-500 Hz 3-axis force “Jerk” metric detects movement reliably within 10 ms
Architecture Radio Communicate hints to neighbors Adapt to neighbor mobility, not just node’s own movement
Hint-Aware Protocols • Bit Rate Adaptation • Topology Maintenance • Access Point Policies • Association • Packet scheduling • Pruning • Vehicular network route selection • And more…
Bit Rate Adaptation 802.11g/a bit rates 6 Mbps 9 Mbps 12 Mbps 18 Mbps 24 Mbps 36 Mbps 48 Mbps 54 Mbps Packet encoded at a particular bit rate Finding the best bit rate to transmit a packet Depends on movement, indoors/outdoors, speed
RapidSample • A frame-based protocol for mobile scenarios 1. When a packet fails, probability that the next few packets at the same bit rate will fail is high • Immediately reduce bit rate on packet loss 2. Coherence time of the channel is a few ms (depends on velocity) • Never retry a failed rate and any rate higher than the failed rate for this period of time
RapidSample 3. If the channel is not degrading, it is probably improving • After a few successes at the current bit rate, sample higher rates that have not recently failed (in the last few milliseconds) • If we are wrong about the channel improving and the sampled higher rate fails, revert to the original rate
RapidSample, when device is moving… Up to 75% higher throughput than SampleRate 25% better thanSNR-based protocols that have been trained
But when static… Up to 30%lower throughput than other schemes
Putting It All Together: Hint-Aware Bitrate Adaptation RapidSamplewhen moving SampleRatewhen static Up to 40%-50% better than all other schemes
Routing in Vehicular Mesh Networks • Longevity of links useful – avoids expensive repairs • Links between nodes (vehicles) heading in the same direction tend to last longer • Use heading, position, and speed to obtain link’s connection time estimate (CTE) metric Large difference in headings predicts short-lived link “V2V” Small difference in headings predicts long-lived link
Empirical Evaluation on Taxi Traces The median link duration in seconds for different intervals of heading differences in degrees (180 indicates nodes headed in opposite directions). Links with similar heading lasted 4 to 5 times longer than the median duration over all links
Take-Away Message • Truly mobile devices will soon be dominant • Variety of mobility modes poses problems for wireless protocols • Sensors on these devices give us a new opportunity to develop network protocols • Protocol architecture using sensor hints can significantly improve MAC, link, network layers