Message In Message MIM A Case for Reordering

Message In Message (MIM): A Case for Reordering Transmissions in Wireless Networks Naveen Santhapuri, Srihari Nelakuditi University of South Carolina Justin Manweiler, Souvik Sen, Romit Roy Choudhury, Kamesh Munagala Duke University 1

Outline n Motivation n Contributions n n n Understanding MIM How transmission order affects spatial reuse? Validate benefits of ordering Design MIM aware scheduling framework Future work on MIM and SIC 2

Collision n Signal of Interest (So. I) successful n n n R 1 When SINR is substantially higher And, So. I arrives earlier than Interference Else, collision Collision R 2 So. I Interference 3

Collisions affect Spatial Reuse n MAC protocols designed to avoid collisions n n 802. 11 physical carrier sensing RTS/CTS Greatly limits spatial reuse 4

PHY Capture n n [Kochut: ICNP 04] Interference may not always cause collisions Possible to decode So. I with higher SINR n If So. I arrives within preamble of Interference Preamble 20 us Interference Signal of Interest n Preamble time small (20 us in 802. 11 a) n Benefits small as well 5

Message In Message (MIM) n While receiving a message n n n Receiver looks for new message preamble Stronger Message extracted while receiving ongoing Message Requires higher SINR than when SOI arrives earlier Preamble Interference Signal of Interest n Capability exists in Atheros chipsets 6

802. 11, Capture, and MIM Preamble a) So. I Interference b) So. I Interference d) c) So. I Interference Successful Reception Yes/No (SINR Threshold needed to avoid collision) 802. 11: Yes (10 d. B) No No PHY Capture: Yes (10 d. B) No MIM: Yes (10 d. B) Yes (20 d. B) Different thresholds based on frame ordering 7

Link Ordering Matters Data AP 1 R 1 Controller AP 2 R 2 Data AP 2 AP 1 20 d. B AP 1 R 1 Data 10 d. B R 2 AP 2 R 2 Data R 1 AP 1 transmissions must start first, In general weaker Followed by staggered transmission from AP 2 stronger receiver can recover signal with MIM Allows weaker link R 1 to lock on to signal at low SINR 8

Measurements Rx Tx Order doesn’t affect delivery 1 2 3 4 5 Interferer Order affects Observe that 802. 11 does not enforce the order and delivery ratio thus fails to exploit concurrency

MIM Capable vs MIM Aware n MIM Capable n n Network card can recover MIM Current MAC does not exploit MIM Appropriate ordering happens by chance MIM Aware MAC n n MAC layer harnesses MIM capability Enforces appropriate ordering of transmissions 10

Goal n n Design MIM aware scheduling that reorders transmissions for improving concurrency Research Questions n n n Does MIM awareness yield significant benefits? What is the bound on improvement? How to effect the appropriate order? 11

Optimality Analysis n Integer Programming Formulation in CPLEX Optimal benefits from MIM significant 12

MIM Aware MAC n Shuffle n n n Centralized MIM-aware scheduling protocol For Enterprise Wireless LANs (EWLAN) Why EWLAN? n Increasingly popular architecture n Realizes potential of MIM Controller AP 1 AP 2 AP 3 13

Shuffle: Assumptions n Dominant downward traffic n Powerful controller, Gigabit Ethernet n n Low latency for scheduling/communication Additive Interference n Total = sum of individual interferences 14

Shuffle: Components n n Rehearsal: Measuring interference relations Packet Scheduler n n Use rehearsal and MIMconstraints Output transmission schedules (ordered) Rehearsal Interference Relations Packet Queue Scheduler n Schedule Executer MIM Constraints Ordered Transmissions 15

Measuring Interference Relationships n Periodic Rehearsals n n n APs transmit probes at base rate Each client replies with RSSI values APs too record RSSI values from clients Controller derives interference map Opportunistic rehearsal n n Piggyback RSSI values in data transmissions Continually refine interference map 16

MIM-Aware Scheduler n Objective n n n With MIM, conflicts are asymmetric n n n Maximize concurrency Avoid starvation Conflict graph methods unsuitable Optimal link scheduling is NP-hard Least conflict greedy heuristic n n n Score links based on asymmetric conflicts Links that prevent other links assigned higher score Compute link order based on ‘lower score first’ 17

Batch Selection & Dispatch Batch P 31 Controller P 22 P 32 AP 3 AP 2 AP 1 P 31 P 13 P 12 P 21 P 13 R 11 R 21 R 12 R 13 R 22 R 31 R 32 18

Schedule Execution P 31 Controller P 22 P 31 P 13 P 21 P 32 AP 1 AP 2 AP 3 P 12 R 11 R 21 R 12 n R 13 R 22 R 31 R 32 APs transmit at specified time n DATA Staggering order: AP 1 -AP 3 -AP 2 19

Evaluation n Qualnet simulations n n Throughput and the effect of Fading Parameters n n n 802. 11 a physical model with MIM PLCP: 20 us Fading: Ricean, varying K factor Wired backbone: 1 Gbps ethernet Controller processing latency: 50 us 20

Duke EWLAN Topologies n Client, AP placement traces used to derive topologies (topo 1, topo 2, etc. ) 21

Throughput Comparision Higher gains with Shuffle Gain with scheduling 22

Increasing AP Density n AP density yields higher benefit from Shuffle 23

Impact of Channel Fading n Better throughput gain at lower (Ricean) fading 24

Related Work n Location aware 802. 11: Infocom 05 n n CMAP: NSDI 08 n n Takes advantage of MIM but not ordering Partially benefits from MIM capable hardware Speculative Scheduling for EWLANs: Mobicom 07 n Doesn’t consider MIM 25

Looking Forward n MIM helps recover if So. I is stronger n What if So. I is weaker than interference? n n MIM cannot help Successive Interference Cancellation 26

Successive Interference Cancellation n SIC can be used to recover weaker So. I n n First extract stronger frame Subtract it from the combination Recover weaker frame from residue Feasibility depends on n Strengths of So. I and Interference 27

Interplay of MIM and SIC n Ordering helps SIC too n n If Interference (I) moderately stronger than So. I n Initiate I first to take advantage of MIM n Decode I If I much stronger than So. I n Initiate So. I before I n So. I characterized better for decoding later 28

fir st , S 1 , S Banod 2 de t h. S 2 , S l co oslo 1, da t st S 2 bl de e co da bl e S 2 later S 1, S 2 decodable Rx S 2 Tx 1 Tx 2 st S 1 fir th Bo S 2 too weak to satisfy SNR S 1 S 2 S 1 not decodable RSS of S 2 Characterization & Cancellation S 1 Later S 1, S 2 decodable S 2 not decodable RSS of S 1 29

Future Work n Shuffle n n Implementing and deploying on a test-bed Integrating upload traffic Comparing with other schemes SIC vs MIM n n Explore Characterization vs. Cancellation Advantage of reordering transmissions 30

Thank you 31

Coping with Fading Loss n Immediate corrective rehearsal n n Controller identifies links suspected of fading Schedules a packet batch only for these APs This is a partial rehearsal n Packets are transmitted in serial order n n n APs and clients unaware, send Data and ACKs Controller updates Interference map from ACK RSSIs 32

Idea to explore with SIC n Power Control to enable SIC n n Suppose SINR threshold is 10 d. B SINR is 1 d. B -60 d. Bm -59 d. Bm Tx 1: tx power = 100 m. W Tx 2 33

Idea to explore with SIC n Power control to enable SIC n n Suppose SINR threshold is 10 d. B SINR is 10 d. B after Tx 1 reduced transmit power -60 d. Bm -70 d. Bm Tx 1: tx power = 10 m. W Tx 2 34

Limitations of Capture n Capture does not help when n So. I arrives after the preamble of interference i. e. Receiver locks on to interference Preamble 20 us Interference Signal of Interest n Preamble time small (20 us in 802. 11 a) n Benefits small as well 35