+ All Categories
Home > Documents > Architecture SDN framework goalsmichelus/poster.pdfITA’17, Asilomar’17! s µ⇤ = argmin µ E "...

Architecture SDN framework goalsmichelus/poster.pdfITA’17, Asilomar’17! s µ⇤ = argmin µ E "...

Date post: 28-May-2020
Category:
Upload: others
View: 0 times
Download: 0 times
Share this document with a friend
1
Nicolò Michelusi ([email protected]), Muddassar Hussain, David Love (Purdue) Anoosheh Heidarzadeh, Alex Sprintson (TAMU) 1 2 3 4 5 6 7 8 9 10 11 12 NSF funded project: NSF EARS CNS-1642982: "Real-time Control of Dense, Mobile, Millimeter Wave Networks Using a Programmable Architecture" (2016-2019), Nicolò Michelusi (PI), David Love (co-PI),James Krogmeier (co-PI), Purdue University; Alex Sprintson (co-PI), TAMU; Chris Anderson (co-PI), USNA CNS-1642982 0 0.2 0.4 0.6 0.8 1 False alarm probability, 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1-P MD worst ( , ) = 0.02 ( , ) = 0.08 ( , ) = 0.14 [1] Mm-wave in mobile, dense environments Architecture Energy efficient beam alignment (BA) v Problem: align tx/rx beams with mobile users; QoS constraints; min power Adaptive BA protocol Fractional search Performance Summary v Mobility may disrupt gains of mm-wave Ø Beam-alignment (BA) bottleneck 1. SDN architecture Ø Real-time control in fast-varying networks Ø X-layer to reduce comm. overhead 2. Adaptive energy efficient BA protocol 3. Neyman-Pearson beam design v Future mobile & dense networks demand a . . High flexibility to address overwhelming ... ..communication overhead 1) We propose a flexible architecture for dynamic network control via SDN 2) We design flexible adaptive beam-alignment protocol to minimize energy cost & support QoS 3) We address beam-design to directly incorporate detection performance v “Fractional search” optimal: v iteratively, in closed form Rate constraint Detectability constraint Coverage constraint 2) Data Communication 1) Beam alignment ITA’17, Asilomar’17 ! φ s μ = arg min μ E " L-1 X k=0 E k +(N - L)P L T slot S 0 # s.t. N - L N W tot log 2 1+ γ P L ! L R, E k = φ s ! k , 8k < L, Z L +! L /2 L -! L /2 S L ()d=1, k k = ( 42 k+1 +4k+1 -1 8k+1 , k+1 > 1 2 , k+1 , 0 < k+1 1 2 L = φ d φ s Energy/radiant to attain rate constraint Energy/radiant for detectability 0 5 10 15 20 25 30 35 40 45 0 0.5 1 R = 1 Mbps R = 10 Mbps R = 100 Mbps R = 1000 Mbps Slot index, k 10 1 10 0 10 1 10 2 10 3 10 4 34 36 38 40 42 Bisection search Fractional search exact Rate (Mbps) Power (dBm) MDP formulation v Adaptive beam alignment protocol as finite . horizon Markov decision process Ø Time interval k=0,…,L-1 Ø State U k : width of uncertainty area Ø Action : beam width Ø Cost : sensing energy Ø Final cost : data comm. energy Ø Transitions: depend on ACK/NACK v Analysis of cost-to-go function & structural . properties φ s ! k φ d U L ! k ! k (U k )= 1 2 f k U k ,f k = 1 - 1 2k+1 + f k Fractional factor NP Beam Design v Hypothesis test in each sensing slot: v Neyman-Pearson detector: v Beam design problem: |s H y| 2 ksk 2 2 H 0 7 H 1 γ min c2C M t P MD (c) s.t. P FA (c) δ , kck 2 2 =1 H 1 : y = p P d H ()cs + w, U (R main c ) H 0 : y = p P d H ()cs + w, U (R sec c ) NP Beam Design Problem v Modified problem via relaxation: v Optimal beam: D λ,, D 1 diag(λ)D H 1 - D 0 diag()D H 0 max z 2R,kck 2 2 =1 z s.t. z Gain(c, ), 82 D main c Gain(c, ) φ(γ , δ ), 82 D sec c Performance [1]. J. Song, J. Choi and D. J. Love, "Codebook design for hybrid beamforming in millimeter wave systems," 2015 IEEE International Conference on Communications (ICC), London, 2015, pp. 1298-1303. 33% v Main-love gain improves . . by decreasing v Less energy dispersion ... . outside main-lobe v Better detection performance . by NP-based design v Up to 30% improvement over . state-of-art [1] φ(γ , δ ) -1.5 -1 -0.5 0 0.5 1 1.5 AoD, 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Gain(c, ) ( , ) = 0.02 ( , ) = 0.08 ( , ) = 0.14 [1] R guard R guard c ? (λ, )= v max (D λ,) kv max (D λ,)k 2 Weighted array response matrix main side dual vars optimized via SDP - : beam-width -U: interval of uncertainty - : energy/radiant ACM mmNets’17 SDN framework goals v Enable ......... . programmability . . at MAC & PHY ... . layers v Enable different .. . per-packet .. ... . behaviors v Define and ..... .. incorporate ..... . wireless primitives Apply Clear Write Metadata Meter Instruction Set Goto Output(1) Group(2) eth(src=) icmpv6(type=) Action List Action Set table: 0 buffer: 0 meter: 0 queue: 0 Action Set eth(src=) eth(dst=) queue(5) output(2) Ethernet VLAN VLAN IPv4 TCP Payload 14B 4B 4B 20B 20B 1000B 1063B src: 00:02:03:04:05:06 dst: 00:20:30:40:50:60 type: VLAN (0x8100) pcp: 0 vid: 1000 type: VLAN (0x8100) pcp: 0 vid: 100 type: IPv4 (0x0800) dscp: 0 ecn: 0 protocol: TCP (6) src: 10.0.0.1 dst: 11.1.1.1 src: 10000 dst: 5060 m / a / c / w / md / g Instructions Choice Arrival Extraction Selection Execution Egress Group
Transcript
Page 1: Architecture SDN framework goalsmichelus/poster.pdfITA’17, Asilomar’17! s µ⇤ = argmin µ E " LX1 k=0 E k +(N L)P LTslot S0 # s.t. N L N Wtot log2 1+ P L! L R, E k = s! k, 8k

Nicolò Michelusi([email protected]),Muddassar Hussain,DavidLove(Purdue)Anoosheh Heidarzadeh,AlexSprintson (TAMU)

1 2 3

4 5 6

7 8 9

10 11 12

NSF funded project: NSF EARS CNS-1642982: "Real-time Control of Dense, Mobile,Millimeter Wave Networks Using a Programmable Architecture" (2016-2019),Nicolò Michelusi (PI), David Love (co-PI),James Krogmeier (co-PI), PurdueUniversity; Alex Sprintson (co-PI), TAMU; Chris Anderson (co-PI), USNA

CNS-1642982

0 0.2 0.4 0.6 0.8 1False alarm probability,

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1-P M

Dw

orst

( , ) = 0.02( , ) = 0.08( , ) = 0.14

[1]

Mm-waveinmobile,denseenvironments

Architecture

Energyefficientbeamalignment(BA)v Problem:aligntx/rx beamswithmobileusers;QoS constraints;minpower

AdaptiveBAprotocol

Fractionalsearch Performance

Summary

v Mobilitymaydisruptgainsofmm-wave

Ø Beam-alignment(BA)bottleneck

1. SDNarchitectureØ Real-timecontrolinfast-varyingnetworksØ X-layertoreducecomm.overhead

2. AdaptiveenergyefficientBAprotocol3. Neyman-Pearsonbeamdesign

v Futuremobile&densenetworksdemanda ..Highflexibilitytoaddressoverwhelming .....communicationoverhead

1) Weproposeaflexible architecturefordynamicnetworkcontrolviaSDN

2) Wedesignflexibleadaptivebeam-alignmentprotocoltominimizeenergycost&supportQoS

3) Weaddressbeam-design todirectlyincorporatedetectionperformance

v “Fractionalsearch”optimal:

v iteratively,inclosedform

Rateconstraint

Detectabilityconstraint

Coverageconstraint2)DataCommunication

1)Beamalignment

ITA’17,Asilomar’17

!

�s

µ⇤= argmin

µE"L�1X

k=0

Ek + (N � L)PLTslot

����S0

#

s.t.

N � L

NW

tot

log

2

✓1 + �

PL

!L

◆� R,

Ek = �s!k, 8k < L,Z ↵L+!L/2

↵L�!L/2SL(✓)d✓ = 1,

⇢k

⇢k =

(4⇢2

k+1+4⇢k+1�1

8⇢k+1, ⇢k+1 > 1

2 ,

⇢k+1, 0 < ⇢k+1 12

⇢L =�d

�s

Energy/radianttoattainrateconstraint

Energy/radiantfordetectability

0 5 10 15 20 25 30 35 40 450

0.5

1R = 1 MbpsR = 10 MbpsR = 100 MbpsR = 1000 Mbps

Slotindex,k

10−1 100 101 102 103 10434

36

38

40

42

Bisection searchFractional search exact

Rate(Mbps)

Power(d

Bm)

MDPformulationv Adaptivebeamalignmentprotocolasfinite. horizonMarkovdecisionprocess

Ø Timeinterval k=0,…,L-1Ø State Uk:widthofuncertaintyareaØ Action :beamwidthØ Cost :sensingenergyØ Finalcost :datacomm.energyØ Transitions:dependonACK/NACK

v Analysisofcost-to-gofunction&structural. properties

�s!k

�dUL

!k

!⇤k(Uk) =

1

2fkUk, fk =

✓1� 1

2⇢k+1

◆+

f kFractio

nalfactor

NPBeamDesignv Hypothesistestineachsensingslot:

v Neyman-Pearsondetector:

v Beamdesignproblem:

|sHy|2

ksk22

H0

7H1

minc2CMt

PMD(c)

s.t. PFA(c) �, kck22 = 1

H1 : y = ↵pPdH(✓)cs+w, ✓ ⇠ U(Rmain

c )

H0 : y = ↵pPdH(✓)cs+w, ✓ ⇠ U(Rsec

c )

NPBeamDesignProblemv Modifiedproblemviarelaxation:

v Optimalbeam:

D�,⌫ , D1 diag(�)DH1 �D0 diag(⌫)D

H0

max

z2R,kck22=1

z

s.t. z Gain(c, ✓), 8✓ 2 Dmainc

Gain(c, ✓) �(�, �), 8✓ 2 Dsecc

Performance

[1].J.Song,J.ChoiandD.J.Love,"Codebookdesignforhybridbeamforminginmillimeterwavesystems," 2015IEEEInternationalConferenceonCommunications(ICC),London,2015,pp.1298-1303.

33%

v Main-lovegainimproves .. bydecreasing

v Lessenergydispersion .... outsidemain-lobe

v Betterdetectionperformance.byNP-baseddesign

v Upto30%improvementover. state-of-art[1]

�(�, �)

-1.5 -1 -0.5 0 0.5 1 1.5AoD,

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Gai

n(c,

)

( , ) = 0.02( , ) = 0.08( , ) = 0.14

[1]Rguard Rguard

c?(�,⌫) =vmax

(D�,⌫)

kvmax

(D�,⌫)k2

Weightedarrayresponsematrix

main sidedualvars

optimizedviaSDP

- :beam-width- U:intervalofuncertainty- :energy/radiant

ACMmmNets’17

SDNframeworkgoalsv Enable ..........programmability..atMAC&PHY ....layers

v Enabledifferent...per-packet ......behaviors

v Defineand .......incorporate ......wireless primitives

ApplyClearWrite

Metadata

Meter

Instruction Set

Goto

Output(1)

Group(2)eth(src=)icmpv6(type=)

Action List

Action Set

table: 0 buffer: 0meter: 0queue: 0

Action Set

eth(src=)eth(dst=)queue(5)output(2)

Ethernet VLAN VLAN IPv4 TCP Payload

14B 4B 4B 20B 20B 1000B1063B

src: 00:02:03:04:05:06dst: 00:20:30:40:50:60type: VLAN (0x8100)

pcp: 0vid: 1000type: VLAN (0x8100)

pcp: 0vid: 100type: IPv4 (0x0800)

dscp: 0ecn: 0protocol: TCP (6)src: 10.0.0.1dst: 11.1.1.1

src: 10000dst: 5060

m / a / c / w / md / gInstructions

ChoiceArrival Extraction Selection Execution Egress

Group

Recommended