DelayCheck: Scheduling Voice Over Multi-hop Multi-channel ...€¦ · Over Multi-hop Multi-channel Wireless Mesh Networks Ashish Chiplunkar CSE, IITB, India Bhaskaran Raman ... Voice

DelayCheck: Scheduling Voice Over Multi-hop Multi-channel

Wireless Mesh Networks

Ashish ChiplunkarCSE, IITB, India

Bhaskaran RamanCSE, IITB, India

Partha DuttaIBM Research, India

Vijay GabaleCSE, IITB, India

SyNerG: Systems and Networks GroupDepartment of CSE, IIT Bombay, India

COMSNETS 2011

Wireless Mesh Networks & Our Focus

Goal: Effective and efficient scheduling of # of voice calls in a mesh network

Wireless mesh networks

(WiFi/WiMAX/802.15.4) for real-time applications

Usage: Voice applications for

community networking, local voice telephony

Lo3: Low cost, Low power, Local voice for developing regions

Voice over Wireless Mesh Networks

● CSMA based MAC gives poor throughput and delay [Raman, Chebrolu ’05].

● Usage of 802.15.4 radios● Low radio capacity (250 Kbps).● CSMA breaks down for 2 simultaneous calls on 802.15.4

links [Gabale, Raman '10].

● TDMA based MAC to support voice calls (e.g. LiT).

● Scheduling transmission over TDMA mesh is an interesting & important problem.

● Scheduling transmission over TDMA mesh is extensively studied.

Are we reinventing the wheel?

End-to-end delay requirement of VoIP packets: 250ms [Hardy ’02]

● 802.15.4 radio with 250Kbps data rate

Parameter WiFi WiMAX 802.15.4

Data Rate &Slot Size

~ 54 – 1 Mbps,~ 50 – 400 us

~ 75 Mbps,~ 50 – 400 us

250 Kbps,~ 6 ms

TDMA Frame ~ 10 – 30 ms ~ 10 – 30 ms ~ 60 – 80 ms

No. of Data Slots 200 – 25 200 – 25 6 - 8

● End-to-end delay requirement of VoIP packets: 150-250ms [Hardy ’02].

~ 54 – 1 Mbps,~ 50 – 400 us

~ 75 Mbps,~ 50 – 400 us

250 Kbps,~ 6 ms

TDMA Frame ~ 10 – 30 ms ~ 10 – 30 ms ~ 60 – 80 msNo. of Data Slots 200 – 25 200 – 25 6 - 8

● End-to-end delay requirement of VoIP packets: 150-250ms [Hardy ’02].

Efficient scheduling algorithm for strict packet-level delay constraint & admission control?

~ 54 – 1 Mbps,~ 50 – 400 us

~ 75 Mbps,~ 50 – 400 us

250 Kbps,~ 6 ms

TDMA Frame ~ 10 – 30 ms ~ 10 – 30 ms ~ 60 – 80 msNo. of Data Slots 200 – 25 200 – 25 6 - 8

Our Requirements

Parameter Algorithm

Optimization goalMax. Calls admitted

Multi-radio, Multi-channel Yes, Yes

Joint routing & scheduling Yes

Admission control based on delay constraint

Comparison with Prior Work

ParameterScheduling &

CAC WiMAX, [Sahoo'10]

Joint channel assign & link scheduling,

[Alicherry'05]

Min. Length TDMA

schedule, [Djukic'09]

Generalized link

activation, [Narlikar'06]

Algorithm

Optimization goalCAC for delay & jitter constraints

Network thrput w.r.t

fairness

Min. End-to-end delay

Max. Link scheduling

Max. Calls admitted

Multi-radio, Multi-channel No, No No, Yes No, No No, Yes Yes, Yes

Joint routing & scheduling No Yes No Yes Yes

Yes No No No Yes

Prior Work and DelayCheck

ParameterScheduling &

CAC WiMAX, [Sahoo'10]

Joint channel assign & link scheduling,

[Alicherry'05]

Min. Length TDMA

schedule, [Djukic'09]

Generalized link

activation, [Narlikar'06]

DelayCheck

Optimization goalCAC for delay & jitter constraints

Network thrput w.r.t

fairness

Min. End-to-end delay

Max. Link scheduling

Max. Calls admitted

Multi-radio, Multi-channel No, No No, Yes No, No No, Yes Yes, Yes

Joint routing & scheduling No Yes No Yes Yes

Yes No No No Yes

DelayCheck: Efficient scheduling for real-time traffic

Delay Constrained Scheduling in WiFi Mesh?

Admission control based on strict delay constraint

802.11 TDMA mesh [Fractel setting], frame

200ms 100 slts, 3 chns, 500 clients.

Prior work:

Calls accepted but delay exceeded!

Delay Constrained Scheduling in WiFi Mesh?

Admission control based on strict delay constraint

DelayCheck:

Delay constrained, + greater no. of calls accepted!

802.11 TDMA mesh [Fractel setting], frame

200ms 100 slts, 3 chns, 500 clients.

Outline of Talk

DelayCheck problem formulation

Delay of a hop in TDMA scheduling

DelayCheck algorithm: 3 phases

Polynomial algorithm when delay is input

Simulation results

Implementation issues

Conclusion

DelayCheck: Problem Formulation

DelayCheck: Scheduler Inputs

# of slots

Src-Dst, delay f(k+1)

Netw, Interf. graph

# of radios, channels

Curr. 'k' flows, sched state

Online Scheduler

Centralized scheduler

Channel conditions do not change with time

Slot-level channel switching

Unsplittable flows (single path for a flow)

Flows arrive over time (online problem)

6 slots in a scheduling interval Schedule repeats

DelayCheck: Scheduler Constraints

Constraints:Primary interference, Secondary interference, End-to-end delay

# of slots

Netw, Interf. graph

Online Scheduler

Link Scheduling

Delay Constraint

Channel Assignment

Route Selection

DelayCheck: Scheduler Output

New sched/ f(k+1) rejected

Objective:Maximize # of calls admitted

Output: Assign a routing path, time slot & channel for each link

# of slots

Netw, Interf. graph

Link Scheduling

Delay Constraint

Channel Assignment

Route Selection

Online Scheduler

Constraints:Primary interference, Secondary interference, End-to-end delay

DelayCheck: Algorithm

Sl(L1)

A B CL1 L2

TDMA frame (Scheduling Interval)Case 1: In order scheduling.

Scheduling Delay = {Sl(L2) – Sl(L1)} + {Sl(L3) - Sl(L2)}

Links on a path for a flow

Direction of packet traversal

Sl(L2) Sl(L3)

Control slot

Data slot

Sl(L1)

A B CL1 L2

TDMA frame (Scheduling Interval)Case 2: Out of order scheduling.

Scheduling Delay = {N – Sl(L1) + Sl(L2)} + {N – Sl(L2) + Sl(L3)}

N slots

Sl(L2) Sl(L3)

Sl(L3) Sl(L2) Sl(L1)

Control slot

Data slot

Sl(L1)

A B CL1 L2

N slots

Sl(L2) Sl(L3)

A packet’s scheduling delay on a hop depends on whether incoming and outgoing links are scheduled in-order or out-of-order [Djukic ’07].

Delay of a path = sum of the scheduling delay of every hop on the path.

Deterministic/TDMA scheduling ensures that a packet faces only scheduling delay but no queuing delay.

Control slot

Data slot

Sl(L3)

A B CL1 L2

Sl(L2) Sl(L1)

Out of order link scheduling spans multiple TDMA frames, but overall end-to-end scheduling delay of a path can still be less than the given delay constraint.

Control slot

Data slot

End-to-end delay < deadline

How Do We Solve The Scheduling Problem?

Joint Routing, Scheduling

Ensuring DelayConstraint

Maximum # of Voice Calls

DelayCheck: Solution Sketch

● Delay constrained scheduling is NP-hard.● DelayCheck is an online heuristic.

● Three phases upon receiving a flow request

1. Construct auxiliary graph G’ from G and already

scheduled flow

• Every path in G’ is a feasible schedule of a given delay (assuming 1-hop/2-hop interference).

2. Use Dijkstra's algorithm to find shortest path in G’.

3. Post processing to tweak the schedules with

interference arising in pathological cases.

G’ is auxiliary graph, G is network connectivity graph

Phase I: Constructing Auxiliary Graph

A BX Y

Direction of flow from X to Y Selected

Graph edge

CDirected auxiliary graph G’F = (X, Y, D). (D = 3 slots.)

A BX Y

(A,1,1,1)

(A,2,1,1)

(A,1,1,2)

(A,2,1,2)X

(B,1,1,2)

(B,2,1,2)

(B,1,1,3)

(B,2,1,3)(A,1,1,3)(A,2,1,3)

(Y,1,1,3)

(Y,2,1,3)

Selected Path

Graph edge

(A,3,1,1)

(A,3,1,2)

(A,3,1,3)

(B,3,1,2)

(B,3,1,3)(Y,3,1,3)

(C,3,1,3)

(C,1,1,3)

(C,2,1,3)

Directed auxiliary graph G’F = (X, Y, D). (D = 3 slots.)

A vertex in G’ is a tuple (v, s, c, d), free slot s, free channel c, d <= D.

(B,1,1,2)

Direction of flow from X to Y

# of slots = 3# of radios = 1# of channels = 1# of delay values = 3

Scheduling interval = TDMA frame

A BX Y

(A,1,1,1)

(A,2,1,1)

(A,1,1,2)

(A,2,1,2)X

(B,1,1,2)

(B,2,1,2)

(B,1,1,3)

(B,2,1,3)(A,1,1,3)(A,2,1,3)

(Y,1,1,3)

(Y,2,1,3)

Selected Path

Graph edge

(A,3,1,1)

(A,3,1,2)

(A,3,1,3)

(B,3,1,2)

(B,3,1,3)(Y,3,1,3)

(C,3,1,3)

(C,1,1,3)

(C,2,1,3)

Link between (v, s, c, d) and (v’, s’, c’, d’) in G provided 4 rules are met.

• Rule 1: v and v’ are neighbors in G’, ensures connectivity in routing path • Rule 2: there are free radios at v and v’, eliminates primary interference • Rule 3: channel c’ in slot s’ free and non-interfering with previous (s, c) link eliminates secondary interference • Rule 4: d’ = d + N – (s’ – s) if s > s’ (out-of-order scheduling, N = #slots in a frame) = d + s’ – s if s < s’ (in-order scheduling), ensures delay constraint

(B,1,1,2)

A BX Y

(A,1,1,1)

(A,2,1,1)

(A,1,1,2)

(A,2,1,2)X

(B,1,1,2)

(B,2,1,2)

(B,1,1,3)

(B,2,1,3)(A,1,1,3)(A,2,1,3)

(Y,1,1,3)

(Y,2,1,3)

Selected Path

Graph edge

(A,3,1,1)

(A,3,1,2)

(A,3,1,3)

(B,3,1,2)

(B,3,1,3)(Y,3,1,3)

(C,3,1,3)

(C,1,1,3)

(C,2,1,3)

• Rule 1: v and v’ are neighbors in G’, ensures connectivity in routing path • Rule 2: there are free radios at v and v’, eliminates primary interference • Rule 3: channel c’ in slot s’ free and non-interfering with previous (s, c) link eliminates secondary interference • Rule 4: d’ = d + N – (s’ – s) if s > s’ (out-of-order scheduling, N = #slots in a frame) = d + s’ – s if s < s’ (in-order scheduling), ensures delay constraint

(B,1,1,2)

A BX Y

(A,1,1,1)

(A,2,1,1)

(A,1,1,2)

(A,2,1,2)X

(B,1,1,2)

(B,2,1,2)

(B,1,1,3)

(B,2,1,3)(A,1,1,3)(A,2,1,3)

(Y,1,1,3)

(Y,2,1,3)

Selected Path

Graph edge

(A,3,1,1)

(A,3,1,2)

(A,3,1,3)

(B,3,1,2)

(B,3,1,3)(Y,3,1,3)

(C,3,1,3)

(C,1,1,3)

(C,2,1,3)

What does a link in G' mean? Say link between w = (A,1,1,1) and w' = (B,2,1,2).

Link (A,B) in G is scheduled In slot 2 On channel 1 Delay of the path so far (from the source) is 2.

Slot 2, chanl 1, delay 2

A BX Y

(A,1,1,1)

(A,2,1,1)

(A,1,1,2)

(A,2,1,2)X

(B,1,1,2)

(B,2,1,2)

(B,1,1,3)

(B,2,1,3)(A,1,1,3)(A,2,1,3)

(Y,1,1,3)

(Y,2,1,3)

Selected Path

Graph edge

(A,3,1,1)

(A,3,1,2)

(A,3,1,3)

(B,3,1,2)

(B,3,1,3)(Y,3,1,3)

(C,3,1,3)

(C,1,1,3)

(C,2,1,3)

Properties: Paths from source to a node of the form (destination, *, *, *)

Links on the path have an interf-free slot and channel assignment. Packet reaches the destination within delay constraint.

Phase II: Apply Dijkstra's Algorithm

A BX Y

(A,1,1,1)

(A,2,1,1)

(A,1,1,2)

(A,2,1,2)X

(B,1,1,2)

(B,2,1,2)

(B,1,1,3)

(B,2,1,3)(A,1,1,3)(A,2,1,3)

(Y,1,1,3)

(Y,2,1,3)

Selected Path

Graph edge

(A,3,1,1)

(A,3,1,2)

(A,3,1,3)

(B,3,1,2)

(B,3,1,3)(Y,3,1,3)

(C,3,1,3)

(C,1,1,3)

(C,2,1,3)

Direction of flow from X to YUse Dijkstra's algorithm to

find a shortest path from src to dst in G'.

Heuristic: Minimizing length of the path minimizes # of resources used.

A resource here is given by (e, s, c): using a slot s with a channel c on a link e.

A BX Y

(A,1,1,1)

(A,2,1,1)

(A,1,1,2)

(A,2,1,2)X

(B,1,1,2)

(B,2,1,2)

(B,1,1,3)

(B,2,1,3)(A,1,1,3)(A,2,1,3)

(Y,1,1,3)

(Y,2,1,3)

Selected Path

Graph edge

(A,3,1,1)

(A,3,1,2)

(A,3,1,3)

(B,3,1,2)

(B,3,1,3)(Y,3,1,3)

(C,3,1,3)

(C,1,1,3)

(C,2,1,3)

X-A A-B B-Y X-A A-B B-Y

Slot 2Ch 1

Delay 1

Slot 3Ch 1

Delay 2

Slot 1Ch 1

Delay 3 TDMA frame

Schedule Schedule repeats

Direction of flow from X to YUse Dijkstra's algorithm to

find a shortest path from src to dst in G'.

A BX Y

(A,1,1,1)

(A,2,1,1)

(A,1,1,2)

(A,2,1,2)X

(B,1,1,2)

(B,2,1,2)

(B,1,1,3)

(B,2,1,3)(A,1,1,3)(A,2,1,3)

(Y,1,1,3)

(Y,2,1,3)

Selected Path

Graph edge

(A,3,1,1)

(A,3,1,2)

(A,3,1,3)

(B,3,1,2)

(B,3,1,3)(Y,3,1,3)

(C,3,1,3)

(C,1,1,3)

(C,2,1,3)

Use Dijkstra's algorithm to find a shortest path from src to dst in G'.

Note 1: Minimizing # resources performs close to optimal in simulation.

Note 2: Algorithm is pseudo-polynomial (complexity O((V x S x C x D)2) – but can be made polynomial by adding weights to edges.

Phase II: Polynomial in D

(B,3,1,2)

(V, S, C, D): 4th value – delay

(Y,1,1,3) delay

(B,3,1,2)

(Y,1,1,m+1)

(V, S, C, M): 4th value – hop

Delay = f (s, s') = 3 – (3 – 1) = 1

(B,3,1,m)

(Y,1,1,3) delay

Dijkstra's algorithm to find “minimum length path which has least delay”. (Complexity O((V x S x C x V)2).

Rule 4 changed: there is link between two nodes (v, s, c, m) and (v', s', c', m+1) in G' with delay (s' – s) mod N.

Formulate and solve real-life constrained fuel problem: minimize toll required to pay, consuming least amount of petrol (with toll/road = 1).

(B,3,1,2)

(Y,1,1,m+1)

(V, S, C, M): 4th value – hop

Delay = f (s, s') = 3 – (3 – 1) = 1

(B,3,1,m)

(Y,1,1,3) delay

Phase III: Post-processing

Circular pond

Nodes placed along the pond

L1 & L

5 : 3 hops away,

but may still interfere.

Detect such interfering links.

Reschedule on different slot or channel.

DelayCheck: Features

● Online algorithm.

● Polynomial time algorithm (heuristic) for delay constrained scheduling.

● Joint routing, channel assignment and scheduling along with strict packet-level delay constraint.

● Low memory and CPU requirements (explain shortly).

First algorithm to have above set of features

Comparing against Optimal Solution

Integer Linear Programming (ILP) formulation for the delay-constrained, multi-hop, multi-radio, multi-channel routing & TDMA scheduling problem.

Offline algorithm which may reschedule existing flows while scheduling given input flow.

Gives upper bound on number of calls accepted by an offline scheduler.

Necessary condition with lower bounds on in-order and out-of-order scheduling

DelayCheck: Simulation Results

Simulation Set Up

Simulation Setup: – 802.15.4 wireless mesh (Lo3) – 2 Km * 2 Km, 25 infrastructure node and 100 clients – 1 radio per node, 4 channels – At every client, exponential distribution of • inter-call duration (mean = 0.5hrs to 2hrs) • call duration (mean = 2min) – TDMA frame of 60ms with 8 data slots of 6ms each – Duration: 12 hrs

Metrics: – Scheduler efficiency – Call rejection rate

DelayCheck scheduler efficiency = 93%

DelayCheck performs close to optimal.

DelayCheck outperforms even-odd scheduler.

Even-odd scheduler efficiency = 69%

Online DelayCheck scheduler performs close to offline scheduler.

Calls rejected by DelayCheck quite close to calls rejected by offline scheduler

Call rejection rate reduced by 2 fold.

Comparing number of calls rejected by DelayCheck and even-odd scheduling.

DelayCheck: Implementation

Implementing In Practice

Storage requirement of graph? E.g. 25 nodes, 100 slots, 10 channels, 10 hops: O(1010).

Observation: Graph need not be stored, connectivity checking can be done on-the-fly in O(1).

Only storage: Cost matrix O(n). ~64KB.

Implementing In Practice

CPU requirement to search the state space? E.g. 25 nodes, 100 slots, 10 channels, 10 hops, 2.2GHz CPU, 2GB RAM: 8 seconds.

Observation: search space can be suitably partitioned, e.g. Highest numbered slot allotted.

During search we consider most likely partition, then go on exploring bigger partitions. Time reduced to ~ 1 second.

Storage requirement of graph? E.g. 25 nodes, 100 slots, 10 channels, 10 hops: O(1010).

Observation: Graph need not be stored, connectivity checking can be done on-the-fly in O(1).

Only storage: Cost matrix O(n). ~64KB.

DelayCheck: Concluding Remarks

Goal: Efficient, delay constrained scheduler to maximize number of voice calls

Joint routing, channel assignment, link scheduling Works in 3 phases Can be extended for 'm'-hop interference model Delay-constrained scheduling

Feasibility & effectiveness via implementation & evaluation Close to optimal behavior 2 x improvement over prior work Low memory & CPU requirement yet being efficient

DelayCheck publication/presentation at:http://www.cse.iitb.ac.in/~vijaygabale/

Vijay Gabalevijaygabale@cse.iitb.ac.in

SyNerG: Systems and Networks Grouphttp://www.cse.iitb.ac.in/synerg

COMSNETS 2011

Thanks! Questions???

This work is funded by:IU-ATC Project, DST, Goverment of India and

UK EPSRC Digital Economy Programme

Phase II: Incorporating 2-hop Interference

(v, s, c, m) (s, c): Check for interference on previous hop

• Rule 3: channel c’ in slot s’ free and non-interfering with previous (s, c) link, eliminates secondary interference

Phase II: Incorporating 2-hop Interference

(v, s, c, m) (s, c): Check for interference on previous hop

• Rule 3: channel c’ in slot s’ free and non-interfering with previous link e:(s, c) and link before e, e1:(s1, c1) , eliminates secondary interference

(v, s, c, s1, c1, m) (s, c, s1, c1): Check for interference on previous two hops in the path

Effect of Multi-channel Scheduling

Using multiple channels can increase call acceptance by 2-fold.

Increase in call acceptance, 5 channels seem sufficient

DelayCheck: Scheduling Voice Over Multi-hop Multi-channel ...€¦ · Over Multi-hop Multi-channel Wireless Mesh Networks Ashish Chiplunkar CSE, IITB, India Bhaskaran Raman ... Voice

Documents

Selection Metrics for Multi-hop Cooperative Relaying

Adaptive Interference Cancellation System for Multi-hop ...

Throughput-Optimal Multi-hop Broadcast...

WSN Multi Hop

One-Hop Out-of-Band Control Planes for Low-Power Multi-Hop.....

Multi-hop Wireless Networks

Constructing Datasets for Multi-hop Reading Comprehension...

Statistical Admission Control in Multi-Hop Cognitive Radio.....

Sustaining Cooperation in Multi-Hop Wireless Networks

PIP: A Connection-Oriented, Multi-Hop, Multi-Channel TDMA...

TCP with Hop-Oriented Network Coding in Multi-Radio Multi

Understanding Congestion Control in Multi-hop Wireless ...

Framing and Schedule Dissemination for Multi-hop … and...

Designing Autonomic Wireless Multi-hop Networks for Delay...

COOPERATIVE DIVERSITY in WIRELESS TRANSMISSION: Multi-Hop

Multi-Hop Versus Overlay Networks