460 likes | 477 Views
Explicit and Implicit Pipelining in Wireless MAC. Nitin Vaidya University of Illinois at Urbana-Champaign nhv@uiuc.edu Joint work with Xue Yang, UIUC. Goal. Improving performance of MAC protocols. IEEE 802.11 MAC. Channel contention resolved using backoff
E N D
Explicit and Implicit Pipeliningin Wireless MAC Nitin Vaidya University of Illinois at Urbana-Champaign nhv@uiuc.edu Joint work with Xue Yang, UIUC
Goal • Improving performance of MAC protocols
IEEE 802.11 MAC • Channel contention resolved using backoff • Nodes choose random backoff interval from [0, CW] • Count down for this interval before transmission • RTS/CTS handshake before transmission of data packet Random backoff RTS/CTS Data Transmission/ACK
Inefficiency of IEEE 802.11 • Backoff interval should be chosen appropriately for efficiency • Backoff interval with 802.11 far from optimum
Observation • Backoff and RTS/CTS handshake are unproductive: • Do not contribute to goodput Unproductive Random backoff RTS/CTS Data Transmission/ACK
Random backoff RTS/CTS Data Transmission/ACK Stage1 Stage2 Pipelining • Two stage pipeline: • Random backoff and RTS/CTS handshake • Data transmission and ACK • “Total” pipelining: Resolve contention completely in stage 1
How to pipeline? • Use two channels • Control Channel: Random backoff and RTS/CTS handshake • Data Channel: Data transmission and ACK Random backoff RTS/CTS Random backoff RTS/CTS Random backoff RTS/CTS Data Transmission/ACK Data Transmission/ACK
Related Work • Todd and Mark, “Capacity allocation in multiple access networks,” IEEE Trans. on Communications, vol. COM-33, 1985. • Todd and Mark considered a similar “total pipelining” model, where the channel contention is completely resolved on a separate scheduling sub-channel. • Results: • If MAC access control overhead consists of bandwidth-dependent component only, not possible to improve throughput by pipelining • If MAC access control overhead consists of bandwidth-independent component only, substantial improvement can be expected.
“Total” pipelining of IEEE 802.11 in wireless networks • IEEE 802.11 contention resolution overhead consists of both bandwidth-independent and bandwidth-dependent components. • Random Backoff Duration is bandwidth-independent • Collision detection and RTS/CTS transmissions are bandwidth-dependent overhead
Random backoff RTS/CTS Random backoff RTS/CTS Random backoff RTS/CTS Pipelining works well only if two stages are balanced! Control Channel Data Transmission/ACK Data Transmission/ACK Data Channel
Difficult to keep the two stages balanced • Length of stage 1 depends on: • The random backoff duration • The number of collisions occurred • Control channel bandwidth • Length of stage 2 depends on: • The data packet size • Data channel bandwidth
How much bandwidth does control channel require? • If small, then • RTS/CTS takes very long time. • Collision detection is slow. • Data Channel may be unnecessarily kept idle due to the lengthened contention resolution duration. • If large, then • The portion of channel bandwidth used for productive data packet transmission is reduced Total bandwidth is fixed!
Difficulties with “Total” Pipelining • The optimum division of channel bandwidth varies with contention level and data packet size • Performance with inappropriate bandwidth division could be even worse than 802.11 DCF
Simulation Results for “Total” PipeliningTotal channel bit rate: 11 Mbps “Total” Pipelining, payload size: 1KB 802. 11 payload size: 1KB Aggregate Throughput (Kbps) “Total” Pipelining, payload size: 512 B 802.11, payload size: 512 B Control Channel Bit Rate (Mbps)
PartialPipeliningModified Two Stage Pipeline • Stage 1: Random backoff phase 1 • Stage 2: Random backoff phase 2, RTS/CTS handshake and Data/ACK transmission Backoff phase 1 Backoff phase 2 RTS/CTS Data/ACK Stage1 Stage2
How to pipeline? • Still use two channels • Narrow Band Busy Tone Channel: Random backoff phase 1 • Data Channel: Random backoff phase 2, RTS/CTS handshake and Data/ACK Stage 1: Busy tone channel Random backoff phase 1 Random backoff phase 1 Random backoff phase 1 Backoff phase 2 RTS/CTS Data/ACK Backoff phase 2 RTS/CTS Data/ACK Stage 2: Data channel
Stage 1 of “Partial Pipelining” • Each station maintains a counter for random backoff phase 1. • The stations, which count to zero first, send a busy tone to claim wining in stage 1, becoming pipelined stations. • Multiple pipelined stations are possible • Other stations are frozen in backoff phase 1 on sensing a busy tone, and are unfrozen when the transmission of the pipelined packet begins.
Stage 2 of “Partial Pipelining” • Only pipelined stations are allowed to contend for the data sub-channel in stage 2. • Stage 2 proceeds quite similar to IEEE 802.11 DCF. • Random backoff phase 2 • RTS/CTS handshake • Data/ACK • Retransmissions upon collisions • Pipelined stations that lose data sub-channel access return back to stage 1.
Sounds like HIPERLAN/1? HIPERLAN / 1 (no pipelining) Elimination Stage Yield Stage Data Transmission Partial Pipelining Random backoff phase 1 Random backoff phase 1 Random backoff phase 1 Backoff phase 2 RTS/CTS Data/ACK Backoff phase 2 RTS/CTS Data/ACK
Gain of “Partial” Pipelining Over “Total” Pipelining • Since there is no need to completely resolve contention in pipelined stage 1, the length of stage 1 is elastic so the two stages can be kept balanced • No packets transmitted on busy tone channel • bandwidth can be small • the difficulty of deciding optimum bandwidth division in “total pipelining” is eased
Benefits of Partial Pipeline Because of pipelining, stages 1 and 2 proceed in parallel. Stage 1 costs little except for a narrow band busy tone channel Partial Pipelining Random backoff phase 1 Random backoff phase 1 Random backoff phase 1 Backoff phase 2 RTS/CTS Data/ACK Backoff phase 2 RTS/CTS Data/ACK
Benefits of Partial Pipeline (cont.) • By migrating most of the backoff (i.e., random backoffphase 1) to the busy tone sub-channel, the bandwidthindependent overhead is reduced. • Cost of backoff = Channel bandwidth * backoff duration • Using IEEE 802.11 DSSS, the backoff duration could be several milliseconds Data Channel Bandwidth Area = cost of backoff Busy Tone Channel Bandwidth Backoff Duration
Benefits of Partial Pipeline (cont.) • Only pipelined stations can contend for the data sub-channel in stage 2 • With large probability, the number of pipelined stations, which contend for the data sub-channel at the same time, is small • reduces collision probability on the data sub-channel • Leading tothe reduced bandwidth-dependent overhead. Stage 1 Stage 2
Performance Results of Partial Pipelining Partial Pipelining Payload size: 512 B Improved Channel Utilization 37.5% improvement 802.11, Payload size: 512 B Normalized Throughput Contending Stations (N)
Performance Results of Partial Pipelining Improved Packet Access Delay 42% improvement 802.11 Payload size: 512 B Average Packet Access Delay (s) Partial Pipelining Payload size: 512 B Contending Stations (N)
Performance Results of Partial Pipelining Improved Access Energy Efficiency 37% improvement 802.11, Payload size: 512 B Joule / Kbps Partial Pipelining Payload size: 512 B Contending Stations (N)
Busy Tone May not Always Be Sensed Partial Pipelining with busy tone detection probability: 100%, 80% and 50% , respectively 802.11 Normalized Throughput Partial Pipelining with 0% busy tone detection probability Contending stations (N)
Observation • With 0% busy tone detection probability: • throughput of partial pipelining degrades due to the increased contention in stage 2 • throughput remains better than 802.11 because of the reduced bandwidth-dependent and bandwidth-independent overhead • This suggests the “implicit” pipelining scheme
Without busy tone in stage 1 • When a station counts its backoff counter to 0, other stations do not know. • Effectively, all stations may count down till the end of stage 1 (as marked by end of ongoing data transmission) • More pipelined stations will appear compared to when using busy tone. • If we can still control the number of pipelined stations within a reasonable range, the benefits of “partial” pipelining may be retained.
Implicit Pipeline • Stage 1: Random backoff phase 1 • Stage 2: Random backoff phase 2, RTS/CTS handshake and Data/ACK transmission Backoff phase 1 Backoff phase 2 RTS/CTS Data/ACK Stage1 Stage2
Still pipeline two stages, but with single channel Random backoff phase 1 Random backoff phase 1 Random backoff phase 1 Implicit stage 1 Backoff phase 2 RTS/CTS Data/ACK Backoff phase 2 RTS/CTS Data/ACK Channelusage
Stage 1 of “Implicit” Pipelining • Each station reduces stage 1 backoff counter by a “reasonable” amount at the end of ongoing data transmission. • “implicit pipelining” • Stations with backoff counter less than zero win stage 1, becoming pipelined stations.
Stage 1 of “Implicit Pipelining” (cont.) • The objective of the backoff algorithm for stage 1 is to keep the number of pipelined stations non-zero but small when ongoing data transmission finishes. • Contention window for stage 1 is increased aggressively (on failure to win data channel access in stage 2) • The more appeared pipelined stations, the more stations that have large contention window sizes – self adjusting. • Backoff counter is decreased faster for stations that have been waiting longer in stage 1
Average number of pipelined stations for “implicit” pipelining Contending stations (N)
Implicit Pipelining • Inherites benefits of “partial pipelining” • Backoff phase 1 proceeds in parallel with other channel activities, leading to the reduced bandwidth-independent channel idle overhead. • Reduces data channel contention by reducing the number of contending stations, leading to the reduced bandwidth-dependent collision overhead.
Implicit Pipelining • Advantages compared with “partial pipelining” • No busy tone channel is needed • Disadvantage compared with “partial pipelining” • More pipelined stations appear from stage 1, leading to the degraded performance in very large networks
Number of Collisions EncounteredPayload size: 512 B, RTS/CTS access method 802.11 Number of RTS Retransmissions Implicit pipelining Partial Pipelining Contending Stations (N)
Performance Results of Implicit Pipelining Implicit pipelining Payload size: 512B Improved Channel Utilization 30% improvement 802.11 Payload size: 512 B Normalized Throughput (Kbps) Contending stations (N)
Performance Results of Implicit Pipelining Improved Packet Access Delay 24% 802.11 Payload size: 512 B improvement Average Packet Access Delay (s) Implicit pipelining Payload size: 512B Contending stations (N)
Performance Results of Implicit Pipelining Improved Access Energy Efficiency 31% 802.11 Payload size: 512 B improvement Joule / Kbps Implicit pipelining Payload size: 512B Contending stations (N)
Performance of Implicit Pipelining in multi-hop Ad hoc networks Throughput ratio of “implicit” pipelining over 802.11 10% - 49% Improvement Aggregate Throughput Ratio Topology Index Simulated in 30 1000m*1000m random networks with 80 contending stations
Performance of Implicit Pipelining in multi-hop Ad hoc networks Packet Access Delay 802.11 11% - 69% Improvement Average Packet Access Delay (s) Implicit pipelining Topology Index Simulated in 30 1000m*1000m random networks with 80 contending stations
Performance of Implicit Pipelining in multi-hop Ad hoc networks Access Energy Efficiency 802.11 9% - 46% Improvement Joule / Kbps Implicit pipelining Topology Index Simulated in 30 1000m*1000m random networks with 80 contending stations
Conclusions • Pipelining can improve MAC performance • “Partial” pipelining is more suitable for wireless networks than “total” pipelining. • Various approaches (e.g., partial pipelining, implicit pipelining) can be conceived for pipelining. • On-going work: Improving pipelining in multi-hop networks
Thanks! Slides last modified February 29, 2004