Supporting VoIP in IEEE Distributed WLANs

Transcription

1 Supporting VoIP in IEEE Distributed WLANs Zuo Liu Supervisor: Dr. Nick Filer July

2 Voice VoIP Applications Constant Streaming Traffic Packetize interval usually ms bytes each packet RTP/UDP Voice Voice Voice Voice Voice Voice Voice Voice Voice Time Sensitive Interactive voice service If round-trip delay is high (e.g. >300 ms), users may suffer in conversation Tolerate to Some Packet Losses Conversation can go on even when some packets are not delivered in time (e.g. 10%) Packet lost concealment techniques 2

3 Existing Research Infrastructure Unbalanced uplink/downlink traffic My Research Fully distributed Contentions for transmission opportunity VoWLAN Scenarios Real World Mixed Internet 3

4 Issues for VoIP in IEEE WLAN 1 Shared Wireless Channel Overhead by contending for transmission opportunity (CSMA/CA) More contentions for VoIP traffic Delay when lost in contention Slot BUSY DIFS Frame SIFS ACK Contention Window Time Traffic Demands 4 Normal (Bursty) Traffic VoIP Traffic Time

5 Issues for VoIP in IEEE WLAN 2 Very Low Efficiency over Wireless Channel Control data overhead is too large Payload is tiny for each voice packet Preamble PLCP Header MAC Header Payload CRC IP+UDP+RTP Headers Data 5

6 micro-seconds (µsec) Estimations Mb/s Mb/s Mb/s Payload Mb/s Overhead 20 11Mb/s 20 54Mb/s b/g (G.711)160 11Mb/s (G.711)160 54Mb/s (G.729)20 11Mb/s (G.729)20 54Mb/s VoIP Call Capacity

7 Buffer Accumulation Traffic Demands VoIP Traffic Time Outgoing Buffer Load tx tx tx tx tx Time 7

8 Delay Accumulation Delay Accumulation Observed from Round-Trip Time Measured in a Real Congested IEEE Distributed WLAN with 12 G.711 calls G.711@54Mb/s Milliseconds Pakcet IDs

9 Round Trip Time in Real WLAN Milliseconds Average RTT Measured by VoIP-RTT-Emulator in Real IEEE802.11b/g Distributed WLAN Calls 4 Calls 6 Calls 8 Calls 10 Calls 12 Calls 14 Calls Number of Simultaneous VoIP Calls 9

10 Packet Delivery Rate within 150 ms Simulated Congestion Test Normalised Packet Delivery Rate within 150 ms Delay Budget after Network Congestion Happens in Simulated IEEE802.11Distributed WLAN 10 Calls 12 Calls 14 Calls 16 Calls Time in Seconds After Network Congestion

11 Auto Cleaning Queue (ACQ) Stale Voice Packets in Outgoing Buffer Not possible being delivered in time Receiver end already played a pseudo sample Wasted transmission opportunity Related Work Dynamic buffer size Drop front/random drop (mainly for TCP) ACQ Actively dropping voice packets older than Giving transmission opportunity to fresh voice packets Compatible with IEEE802.11e Voice (AC_VO) queue Fresh Packet Fresh Packet Fresh Packet Stale Packet Stale Packet 11

12 ACQ Evaluation 1 Milliseconds Average Transmission Latency for VoIP Traffic over Simulated IEEE WLAN G.711@54Mb/s G.729@54Mb/s G.711@54Mb/s + ACQ@100ms budget G.729@54Mb/s + ACQ@100ms budget G.711@54Mb/s + ACQ@150ms budget G.729@54Mb/s + ACQ@150ms budget G.711@54Mb/s + ACQ@200ms budget G.729@54Mb/s + ACQ@200ms budget Calls 4 Calls 6 Calls 8 Calls 10 Calls 12 Calls 14 Calls 16 Calls Number of Simultaneous VoIP Calls 12

13 ACQ Evaluation 2 Packet Delivery Ratio within 150 ms Normalised Packet Delivery Ratio within 150 Milliseconds Delay Budget for VoIP Calls in Simulated IEEE Distributed WLAN G.711@54Mb/s G.729@54Mb/s G.711@54Mb/s + ACQ@100ms budget G.729@54Mb/s + ACQ@100ms budget G.711@54Mb/s + ACQ@150ms budget G.729@54Mb/s + ACQ@150ms budget G.711@54Mb/s + ACQ@200ms budget G.729@54Mb/s + ACQ@200ms budget 2 Calls 4 Calls 6 Calls 8 Calls 10 Calls 12 Calls 14 Calls 16 Calls Number of Simultaneous VoIP Calls

14 Small Packet Aggregation for Wireless Networks Routing Protocols NET De-aggregator Aggregator Out-going Buffer LLC IEEE MAC MAC PHY IEEE PHY 14

15 SPAWN Packet En-queue Process Packet In Check rx Address Block for rx Exists in Queue? N Y Check Block Size Create New Block Y Over Max Payload? N Add to Existing Block 15

16 SPAWN Outgoing Buffer Structure In Out Block3 Block2 Block1 Node B pkt 6 Node B pkt 2 Node A pkt 1 pkt 3 pkt 4 pkt 5 16

17 SPAWN Block Structure Compatible with IEEE No modification on frame header and structure No extra operations required below LLC layer IEEE802.11n aggregation (A- MPDU) resulting in new MAC protocol Flexible for Efficiency Variable number of packets Not forced to reach the maximum payload size MAC Packet Payload MAC Header Block Offset 2 Bytes IP Packet 1 Block Offset 2 Bytes IP Packet 2 CRC 4 Bytes 17

18 SPAWN Evaluation 1 Avg Transmission Latency (Sec) Average Transmission Latency for VoIP Traffic over Simulated IEEE WLAN G.711 using Normal IEEE G.711 using IEEE aggregation G.729 using Normal IEEE G.729 using IEEE aggregation Number of VoIP Conversations 18

19 SPAWN Evaluation 2 1 Normalised Packet Delivery Ratio within 150 Milliseconds Delay Budget for VoIP Calls in Simulated IEEE Distributed WLAN Delivery Rate in 150 ms G.711 using Normal IEEE G.711 using IEEE aggregation G.729 using Normal IEEE G.729 using IEEE aggregation Number of VoIP Conversations

20 Outcomes Conclusions Similar behaviour for VoIP over distributed and infrastructure WLANs Delay accumulation after congestion ACQ resolves the delay accumulation SPAWN effectively increases the network efficiency Compatible with IEEE802.11a/b/g MAC and IEEE802.11e Future Work ACQ and SPAWN in real devices (seeking partners) Dynamic for ACQ Multicasting SPAWN Multi hops, multi flows, multi traffic types, etc. 20

21 Thank you! & Questions? 21