WiFi-Nano: Reclaiming WiFi Efficiency through 800 ns Slots

Transcription

1 WiFi-Nano: Reclaiming WiFi Efficiency through 800 ns Slots Eugenio Magistretti Krishna Kant Chintalapudi * Božidar Radunović # Ramachandran Ramjee * Rice University * Microsoft Research India # Microsoft Research Cambridge

2 Problem Overview [Mbps] ~55 Mbps a/b g n ac/ad data-rates have increased from 1 Mbps to 1 Gbps Throughput performance has not seen a commensurate increase

3 Contribution WiFi-Nano increases throughput up to 100%

4 Why Throughput << Data-rate? Medium Access Preamble Data ACK 54 Mbps DIFS 101.5µs 20µs 224µs SIFS 44µs µs Medium Access Preamble Data ACK 300 Mbps DIFS 101.5µs 32µs 40µs SIFS 36µs µs Medium Access Preamble Data ACK 600 Mbps DIFS 101.5µs 40µs 20µs SIFS 44µs µs

5 Why Throughput << Data-rate? 54 Mbps DIFS Medium Access 101.5µs Preamble Overhead ~45% Data 20µs 224µs SIFS ACK 60µs µs 300 Mbps Medium Access 101.5µs Preamble Data 32µs 40µs SIFS ACK 52µs Overhead ~82% µs 600 Mbps Medium Access Overhead Preamble Data ~91% ACK 101.5µs 40µs 20µs 60µs SIFS µs overhead does not scale with data-rate

6 Motivation Overhead Components SIFS 16% Preambles 40% Communications Overhead Cannot be removed! ACK 2% Slot Duration 42% Single Link Case Medium Access Preamble Data Preamble+ACK DIFS 101.5µs 40µs 20µs SIFS 44µs

7 Motivation Preambles 34% Overhead Components SIFS 14% ACK 2% Slot Duration 23% Collisions 27% Multiple Links Case (30)

8 Objective SIFS 14% Preambles 34% ACK 2% Slot Duration 23% Collisions 27% Reduce slot duration q and reduce the occurrence of collisions q while preserving fairness Remove SIFS

9 Challenge: Slot Duration A B a/n slot duration of 9 µs is close to the minimum feasible 0 1 PropagaDon Clear Channel Assessment (CCA) 4 µs Slot û A 3 B 1 PropagaDon Clear Channel Assessment (CCA) ü Rx/Tx Hardware Turnaround 4 µs 4 µs 0 Key assumption: Preamble detection and transmission are serial

10 Speculative Preambles Preamble detection and transmission occur in parallel Clear Channel Assessment may take multiple slots The slot time can be reduced to a round-trip propagation delay Slot Time (800 ns)

11 Speculative Preambles Slot Time (800 ns) 1. As soon as the backoff expires, a node transmits its preamble 2. CCA: A node transmitting a preamble continues to attempt to detect incoming preambles 3. A node aborts its transmission if it detects a preceding preamble Node A 1 Node B Node C Preamble û Preamble Preamble û Medium Access time decreases from µs to 7.6 µs

12 WiFi-Nano Design Objectives Techniques ü Slot Time Duration Speculative Preambles Collisions Probabilistic Collision Resolution Fairness Counter Roll-back Minimum Slot Size SIFS Speculative ACKs

13 Implementation Detect preambles and their starting time under interference [PN1] Lattice Correlator PN1 PN2 PN3 [PN2] [PN3] Simultaneously transmit and detect preambles Analog Self-Interference Canceller Rx Δ Δ Δ [PN1,PN2] + Δ [PN2,PN3] + QH Noise Canceller Δ Tx Analog Radio Front-end [PN1, PN2,PN3] + Interference may require longer preambles

14 Results Experiments q q q Reliability of Preamble Detection Efficiency Gain and Analysis Fairness Lyrtech Qualnet

15 Preamble Detection Can nodes reliably detect preambles despite self-interference? Slightly longer preambles permit to maintain reliability

16 Efficiency Efficiency f(data rate, #nodes) WiFi-Nano increases the throughput up to 100%

17 Efficiency Efficiency f(data rate, slot time) WiFi-Nano.11a/n.11b WiFi-Nano increases the throughput up to 100%

18 How to Achieve More? Frame Aggregation Works only for single-link bulk downloads Practically difficult to achieve q q q Small packets (TPC ACKs) Short flows (HTTP) Delay sensitive applications (TPC, VoIP)

19 Frame Aggregation Practically difficult to achieve Related work reports 18 kb as average aggregation At 18 kb, WiFi-Nano gains 25% over at 600 Mbps

20 Summary WiFi-Nano permits to q overhead can be > 90% Reduce the slot time to 800 ns q Reduce the occurrence of collisions to nearly 0 q Remove SIFS WiFi-Nano increases throughput up to 100%

21 WiFi-Nano: Reclaiming WiFi Efficiency through 800 ns Slots Q&A Eugenio Magistretti Krishna Kant Chintalapudi * Božidar Radunović # Ramachandran Ramjee * Rice University * Microsoft Research India # Microsoft Research Cambridge