trading coordinat ion for randomness
play

Trading Coordinat ion For Randomness Szymon Chachulski Mike J - PDF document

Trading Coordinat ion For Randomness Szymon Chachulski Mike J ennings, Sachin Kat t i, and Dina Kat abi Wireless mesh net works have high loss rat es Roofnet Avg. 30% loss Obj ect ive: Obj ect ive: High t hroughput despit e lossy links


  1. Trading Coordinat ion For Randomness Szymon Chachulski Mike J ennings, Sachin Kat t i, and Dina Kat abi Wireless mesh net works have high loss rat es Roofnet Avg. 30% loss Obj ect ive: Obj ect ive: High t hroughput despit e lossy links High t hroughput despit e lossy links 1

  2. Use Opport unist ic Rout ing Use Opport unist ic Rout ing R1 0% % 0 5 0% R2 50% dst src % 0 50% R3 50% % 0 R4 • Best single pat h � loss prob. 50% 2

  3. Use Opport unist ic Rout ing R1 0% % 0 5 0% R2 50% dst src % 0 50% R3 50% % 0 R4 • Best single pat h � loss prob. 50% Opport unist ic rout ing promises large Opport unist ic rout ing promises large • I n opp. rout ing [ExOR’05], any rout er t hat hears increase in t hroughput increase in t hroughput loss prob. 0.5 4 = 6% t he packet can f orward it � But Overlap in received packet s � Rout ers f orward duplicat es 3

  4. But Overlap in received packet s � Rout ers f orward duplicat es P 1 P 2 R1 P 1 src dst P 2 P 10 R2 P 1 P 2 But Overlap in received packet s � Rout ers f orward duplicat es R1 P 1 P 1 P 2 src P 2 dst P 1 P 2 P 10 R2 St at e-of -t he-art opp. rout ing, ExOR imposes a global scheduler: - Requires f ull coordinat ion; every node must know who received what - Only one node t ransmit s at a t ime, ot hers list en 4

  5. Global Scheduling? dst src • Global coordinat ion is t oo hard • One t ransmit t er Global Scheduling? dst src • Global coordinat ion is t oo hard Does opport unist ic rout ing Does opport unist ic rout ing • One t ransmit t er � You lost spat ial reuse! have t o be so complicat ed? have t o be so complicat ed? 5

  6. Our Cont ribut ions • Opport unist ic rout ing wit h no global scheduler and no coordinat ion • We use random net work coding • Experiment s show t hat randomness out perf orms bot h current rout ing and ExOR Go Random Each rout er f orwards random combinat ions of packet s 6

  7. Go Random Each rout er f orwards random combinat ions of packet s P 1 R1 α P 1 + ß P 2 P 2 src dst P 1 γ P 1 + δ P 2 R2 P 2 Randomness prevent s duplicat es No scheduler; No coordinat ion Simple and exploit s spat ial reuse Random Coding Benef it s Mult icast P 1 P 2 P 3 src P 4 dst1 dst2 dst3 P1 P1 P 1 P 2 P 2 P2 P3 P 3 P 3 P4 P4 P 4 Wit hout coding � source ret ransmit s all 4 packet s 7

  8. Random Coding Benef it s Mult icast Random combinat ions P 1 8 P 1 + 5 P 2 + P 3 + 3 P 4 P 2 P 3 src 7 P 1 + 3 P 2 + 6 P 3 + P 4 P 4 dst1 dst2 dst3 P1 P1 P 1 P 2 P 2 P2 P3 P 3 P 3 P4 P4 P 4 Wit hout coding � source ret ransmit s all 4 packet s Random coding is more ef f icient t han Random coding is more ef f icient t han Wit h random coding � global coordinat ion 2 packet s are suf f icient global coordinat ion MORE 8

  9. MORE • Source sends packet s in bat ches • Forwarders keep all heard packet s in a buf f er • Nodes t ransmit linear combinat ions of buf f ered packet s a + b + c P3 = P1 P2 a,b,c src A B dst Can comput e linear combinat ions and Can comput e linear combinat ions and P1 4,1,3 4,1,3 P2 sust ain high t hroughput ! 0,2,1 sust ain high t hroughput ! P3 4 + 1 + 3 P3 = P1 P2 4,1,3 0 + 2 + 1 P3 = P1 P2 0,2,1 MORE • Source sends packet s in bat ches • Forwarders keep all heard packet s in a buf f er • Nodes t ransmit linear combinat ions of buf f ered packet s a + b + c P3 = P1 P2 a,b,c src A B dst P1 4,1,3 4,1,3 8,4,7 P2 0,2,1 8,4,7 P3 2 + 1 0,2,1 4,1,3 = 8,4,7 9

  10. MORE • Source sends packet s in bat ches • Forwarders keep all heard packet s in a buf f er • Nodes t ransmit linear combinat ions of buf f ered packet s • Dest inat ion decodes once it receives enough combinat ions o Say bat ch is 3 packet s 1 + 3 + 2 P3 = 1,3,2 P1 P2 5 + 4 + 5 P3 = 5,4,5 P1 P2 4 + 5 + 5 P3 = P1 P2 4,5,5 • Dest inat ion acks bat ch, and source moves t o next bat ch But How Do We Get t he Most Throughput ? • Naïve approach t ransmit s whenever 802.11 allows A dst I f A and B have same inf ormat ion, it is more B ef f icient f or B t o send it Need a Met hod t o Our Madness Need a Met hod t o Our Madness 10

  11. Probabilist ic Forwarding A dst B Probabilist ic Forwarding e1 A e2 dst Loss rat e 0% B Src Loss rat e 50% e1 P1 P2 11

  12. Probabilist ic Forwarding How many packet s 50% of buf f er should I f orward? e1 A e2 dst ? B Src e1 e1 e1 P1 P2 Probabilist ic Forwarding Pr = 0.5 e1 A e2 dst 0% Pr = 1 B 50% Src e1 e1 P1 P2 Comput e f orwarding probabilit ies wit hout Comput e f orwarding probabilit ies wit hout coordinat ion using loss rat es coordinat ion using loss rat es 12

  13. Can ExOR Use Probabilist ic Forwarding To Remove Coordinat ion? Pr = 0.5 Probabilit y of duplicat es is 50% P1 P1 A P2 dst Pr = 1 B • Wit hout random coding � need t o know t he • Wit hout random coding � need t o know t he exact packet s t o f orward every t ime P1 P1 exact packet s t o f orward every t ime • Wit h random coding � need t o know only t he • Wit h random coding � need t o know only t he average amount of overlap average amount of overlap Long-t erm averages are great , but … Wireless is unpredict able over short t ime-scales There are known knowns. These are t hings we know t hat we know. There are known unknowns. That is t o say, t here are t hings t hat we know we don' t know. But t here are also unknown unknowns. MORE needs t o adapt t o short -t erm dynamics 13

  14. Adapt ing t o Short -t erm Dynamics • Need t o balance sent inf ormat ion wit h received inf ormat ion • MORE t riggers t ransmission by recept ions • A node has a credit count er o Upon recept ion, increment t he count er using f orwarding probabilit ies o Upon t ransmission, decrement t he count er • Source st ops � No t riggers � Flow is done Perf ormance 14

  15. Experiment al Set up • We implement ed MORE in Linux • 20-node t est bed • Compare MORE wit h: o Current Rout ing (Single Best Pat h) o ExOR (St at e-of -t he-art Opport unit ic Rout ing) • Experiment o Random source-dest inat ion pairs o Transmit 5 MB f ile Test bed • 20-node t est bed over t hree f loors 15

  16. Test bed • 20-node t est bed over t hree f loors Avg. loss 27% Does MORE I mprove Wireless Throughput ? Avg. Throughput over 180 src-dst pairs [pkts/s] MORE 80 ExOR Current 40 MORE’s t hroughput is MORE’s t hroughput is • 2x bet t er t han current rout ing • 2x bet t er t han current rout ing • 22% bet t er t han ExOR • 22% bet t er t han ExOR 16

  17. Throughput of All Source-Dest inat ion Pairs CDF of 180 source-destination pairs 1 0.8 0.6 Current 0.4 ExOR MORE 0.2 0 0 50 100 150 200 Throughput [packets/s] 0.6 Zoom in on t he worst 10% 0.5 0.4 0.3 Current 0.2 ExOR MORE 4x 0.1 0 0 50 100 MORE addresses dead spot s MORE addresses dead spot s Throughput [packets/s] 17

  18. Sensit ivit y t o Bat ch Size ExOR CDF MORE CDF 1 1 0.75 0.75 0.5 0.5 Batch = 8 pkts Batch = 8 pkts 0.25 0.25 Batch = 128 pkts Batch = 128 pkts 0 0 0 50 100 150 0 50 100 150 Throughput [packets/s] Throughput [packets/s] MORE works f or short f lows MORE works f or short f lows What About Mult icast ? Avg. Throughput Per Destination [pkts/s] Current +200% MORE +260% 100 +320% 50 MORE improves bot h mult icast and unicast MORE improves bot h mult icast and unicast 2 dest inat ions 3 dest inat ions 4 dest inat ions t hroughput t hroughput 18

  19. MORE f or Less! • Lesser coordinat ion and lesser rigidit y o No scheduler • More f lexibilit y o Works on t op of 802.11 � enj oy spat ial reuse o One f ramework f or unicast and mult icast • More t hroughput o 22% bet t er t han ExOR o 2x bet t er t han current rout ing 19

Recommend


More recommend