I-Hong Hou P.R. Kumar
Post on 23-Jan-2016
12 Views
Preview:
DESCRIPTION
Transcript
Admission Control and Scheduling for QoS Guarantees for Variable-Bit-Rate Applications on Wireless Channels
I-Hong Hou
P.R. Kumar
University of Illinois,
Urbana-Champaign
Background: Wireless Networks
There will be increasing use of wireless networks for serving traffic with QoS constraints:
VoIP
Video Streaming
Real-time Monitoring
Networked Control
1/30
Challenges Wireless Network limitation
Non-homogeneous, unreliable wireless links Client QoS requirements
Specified traffic pattern Delay bound Delivery ratio bound Throughput bound
System perspective Fulfill clients with different QoS requirements
2/30
Goal of the Paper Prior work [Hou, Borkar, and Kumar]:
All clients generate traffic with the same rate Admission control and packet scheduling policies
Q: How to deal with more complicated traffic patterns? Applications with variable-bit-rate (VBR) traffic
MPEG streaming Clients generate traffic with different rates
This work extends results to arbitrary traffic patterns
3/30
Client-Server Model A system with N wireless clients and one AP Time is slotted One packet transmission in each slot AP schedules all transmissions
4/30
AP1
2
slot length = transmission duration
3
Channel Model Unreliable, non-homogeneous wireless channels
successful with probability pn
failed with probability 1-pn
p1,p2,…,pN may be different
5/30
AP1
2p1p2
3
p3
Uplink Protocol Poll (ex. CF-POLL in 802.11 PCF) Data No need for ACK pn = Prob( both Poll/Data are delivered)
6/30
AP1
2p1p2POLL
Data
3
p3
Downlink Protocol Data ACK pn = Prob( both Data/ACK are delivered)
7/30
AP1
2p1p2Data
ACK
3
p3
Traffic Model Group time slots into intervals with τ time slots Clients may generate packets at the beginning of
each interval
8/30
AP1
2
3
p1p2
p3
τ{1,.,3}
{1,.,3}
{1,.,3} {.,2,.}
{.,2,.}
{.,2,.}
{1,2,3}
{1,2,3}
{1,2,3}
Delay Bound Deadline = Interval Packets are dropped if not delivered by the deadline Delay of successful delivered packet is at most τ
9/30
AP1
2
3
p1p2
p3
{1,.,3}
{1,.,3}
{1,.,3} {.,2,.}
{.,2,.}
{.,2,.}
{1,2,3}
{1,2,3}
{1,2,3}
τ
arrival deadline
S I
Packet Scheduling
10/30
AP1
2
3
p1p2
p3
SF
F
I
forced idleness{1,.,3}
{1,.,3}
{1,.,3} {.,2,.}
{.,2,.}
{.,2,.}
{1,2,3}
{1,2,3}
{1,2,3}
dropped
S I
Timely Throughput Timely throughput = avg. # of
delivered packets per interval
11/30
AP1
2
3
p1p2
p3
SF
F
I
Client # Throughput
1 0
2 0.5
3 0.5
{1,.,3}
{1,.,3}
{1,.,3} {.,2,.}
{.,2,.}
{.,2,.}
{1,2,3}
{1,2,3}
{1,2,3}
S I
Packet Arrivals Distribution of packet
arrivals is specified
12/30
AP1
2
3
p1p2
p3
SF
F
I
{1,.,3}
{1,.,3}
{1,.,3} {.,2,.}
{.,2,.}
{.,2,.}
{1,2,3}
{1,2,3}
{1,2,3}
Arrival Proportion of Occurrences
{1,3} 1/3
{2} 1/3
{1,2,3} 1/3
S I
QoS Requirements Client n requires timely throughput qn
Delivery ratio requirement of client n
= qn /{arrival prob. of client n}
13/30
AP1
2
3
p1p2
p3
SF
F
I
{1,.,3}
{1,.,3}
{1,.,3} {.,2,.}
{.,2,.}
{.,2,.}
{1,2,3}
{1,2,3}
{1,2,3}
Client # Delivery ratio
1 0
2 1
3 1
Problem Formulation Admission control
Given τ, packet arrivals, pn, qn, decide whether a set of clients is feasible
Scheduling policy Design a policy that fulfills every feasible set of
clients
14/30
The proportion of time slots needed for client n is
Work Load
1 nn
n
qw
p
15/30
The proportion of time slots needed for client n is
Work Load
1 nn
n
qw
p
15/30
expected number of time slots needed for a successful transmission
The proportion of time slots needed for client n is
Work Load
1 nn
n
qw
p
15/30
number of required successful transmissions in an interval
The proportion of time slots needed for client n is
Work Load
1 nn
n
qw
p
15/30
normalize by interval length
The proportion of time slots needed for client n is
We call wn the “work load”
Work Load
15/30
1 nn
n
qw
p
S I
Necessary Condition for Feasibility Necessary condition from classical queuing theory: But the condition is not sufficient Packet drops by deadline misses cause more idleness than in
queuing theory
16/30
AP1
2
3
p1p2
p3
SF
F
I
11
N
nnw
{1,.,3}
{1,.,3}
{1,.,3} {.,2,.}
{.,2,.}
{.,2,.}
{1,2,3}
{1,2,3}
{1,2,3}
Stronger Necessary Condition Let IS = Expected proportion of the idle time when
the server only works on S IS decreases as S increases
Theorem: the condition is both necessary and sufficient
Admission control checks the condition
1, {1,2,..., }n Sn S
w I S N
17/30
Largest Debt First Scheduling Policies
Give higher priority to client with higher “debt”
18/30
AP1
2
3
p1p2
p3
{1,2,3}
{1,2,3}
{1,2,3}F F S
F S
F
Two Definitions of Debt The time debt of client n
time debt = wn – actual proportion of transmission time given to client n
The weighted delivery debt of client n weighted delivery debt = (qn – actual timely throughput)/pn
Theorem: Both largest debt first policies fulfill every feasible set of clients Feasibility Optimal Policies
19/30
Evaluation Methodology
Evaluate five policies: DCF Enhanced DCF (EDCF) by 802.11e PCF with randomly assigned priorities (random) Time debt first policy Weighted-delivery debt first policy
Metric: Shortfall in Timely Throughput
20/30
Evaluated Applications VoIP
Generate packets periodically Duplex traffic Clients may generate packets by different period
MPEG Generate packets probabilistically Only downstream traffic Clients may generate packets by different probability
21/30
VoIP Traffic ITU-T G.729.1
Bit rates between 8 kb/s to 32 kb/s Different bit rates correspond to different periods
8kb/s – 32 kb/s bit rates 20 ms interval length
160 Byte packet 11 Mb/s transmission rate
610 μs time slot 32 time slots in an interval
22/30
VoIP Clients Two groups of clients:
Feasible set: 6 group A clients, 5 group B clients Infeasible set: 6 group A clients, 6 group B clients
Group A Group B
60 ms (3 intervals) period 40 ms (2 intervals) period
21.3 kb/s traffic 32 kb/s traffic
require 99% delivery ratio require 80% delivery ratio
Starting times evenly spaced
Channel reliabilities range from 61% to 67%
23/30
VoIP Results: A Feasible Set
24/30
0
1
2
3
4
5
0 1 2 3 4 5 6
Time (sec)
Sh
ort
fall
in T
imel
y T
hro
ug
hp
ut Weighted-
delivery
0
1
2
3
4
5
0 1 2 3 4 5 6
Time (sec)
Sh
ort
fall
in T
imel
y T
hro
ug
hp
ut Weighted-
delivery
Time-based
VoIP Results: A Feasible Set
fulfilled
24/30
VoIP Results: A Feasible Set
24/30
0
1
2
3
4
5
0 1 2 3 4 5 6
Time (sec)
Sh
ort
fall
in T
imel
y T
hro
ug
hp
ut Weighted-
delivery
Time-based
Random
VoIP Results: A Feasible Set
24/30
0
1
2
3
4
5
0 1 2 3 4 5 6
Time (sec)
Sh
ort
fall
in T
imel
y T
hro
ug
hp
ut Weighted-
delivery
Time-based
Random
DCF
VoIP Results: A Feasible Set
24/30
0
1
2
3
4
5
0 1 2 3 4 5 6
Time (sec)
Sh
ort
fall
in T
imel
y T
hro
ug
hp
ut Weighted-
delivery
Time-based
Random
DCF
EDCF
VoIP Results: An Infeasible Set
25/30
0
1
2
3
4
5
6
0 1 2 3 4 5 6
Time (sec)
Sh
ort
fall
in T
imel
y T
hro
ug
hp
ut
Weighted-delivery
0
1
2
3
4
5
6
0 1 2 3 4 5 6
Time (sec)
Sh
ort
fall
in T
imel
y T
hro
ug
hp
ut
Weighted-delivery
Time-based
VoIP Results: An Infeasible Set
small shortfall
25/30
VoIP Results: An Infeasible Set
25/30
0
1
2
3
4
5
6
0 1 2 3 4 5 6
Time (sec)
Sh
ort
fall
in T
imel
y T
hro
ug
hp
ut
Weighted-delivery
Time-based
Random
VoIP Results: An Infeasible Set
25/30
0
1
2
3
4
5
6
0 1 2 3 4 5 6
Time (sec)
Sh
ort
fall
in T
imel
y T
hro
ug
hp
ut
Weighted-delivery
Time-based
Random
DCF
VoIP Results: An Infeasible Set
25/30
0
1
2
3
4
5
6
0 1 2 3 4 5 6
Time (sec)
Sh
ort
fall
in T
imel
y T
hro
ug
hp
ut
Weighted-delivery
Time-based
Random
DCF
EDCF
MPEG Traffic Model MPEG VBR traffic by a Markov chain
consisting of three activity states (Martin et al)
MAC: 802.11a
6 ms interval length 1500 Bytes packet
54 Mb/s transmission rate 9 time slots in an interval
Activity Great High Regular
Arrival probability 1 0.8 0.75
26/30
MPEG Clients Two groups of clients
Group A generates traffic according to Martin et al and requires 90% delivery ratio
Group B generates traffic half as often as A and requires 80% delivery ratio
The nth client in each group has (60+n)% channel reliability
Feasible set: 4 group A clients, 4 group B clients Infeasible set: 5 group A clients, 4 group B clients
27/30
MPEG Results: A Feasible Set
28/30
0
0.2
0.4
0.6
0.8
0 1 2 3 4 5 6
Time (sec)
Sh
ort
fall
in T
imel
y T
hro
ug
hp
ut Weighted-
delivery
0
0.2
0.4
0.6
0.8
0 1 2 3 4 5 6
Time (sec)
Sh
ort
fall
in T
imel
y T
hro
ug
hp
ut Weighted-
delivery
Time-based
MPEG Results: A Feasible Set
fulfilled
28/30
MPEG Results: A Feasible Set
28/30
0
0.2
0.4
0.6
0.8
0 1 2 3 4 5 6
Time (sec)
Sh
ort
fall
in T
imel
y T
hro
ug
hp
ut Weighted-
delivery
Time-based
Random
MPEG Results: A Feasible Set
28/30
0
0.2
0.4
0.6
0.8
0 1 2 3 4 5 6
Time (sec)
Sh
ort
fall
in T
imel
y T
hro
ug
hp
ut Weighted-
delivery
Time-based
Random
DCF
MPEG Results: A Feasible Set
28/30
0
0.2
0.4
0.6
0.8
0 1 2 3 4 5 6
Time (sec)
Sh
ort
fall
in T
imel
y T
hro
ug
hp
ut Weighted-
delivery
Time-based
Random
DCF
EDCF
MPEG Results: An Infeasible Set
29/30
0
0.2
0.4
0.6
0.8
1
1.2
0 1 2 3 4 5 6
Time (sec)
Sh
ort
fall
in T
imel
y T
hro
ug
hp
ut
Weighted-delivery
0
0.2
0.4
0.6
0.8
1
1.2
0 1 2 3 4 5 6
Time (sec)
Sh
ort
fall
in T
imel
y T
hro
ug
hp
ut
Weighted-delivery
Time-based
MPEG Results: An Infeasible Set
small shortfall
29/30
MPEG Results: An Infeasible Set
29/30
0
0.2
0.4
0.6
0.8
1
1.2
0 1 2 3 4 5 6
Time (sec)
Sh
ort
fall
in T
imel
y T
hro
ug
hp
ut
Weighted-delivery
Time-based
Random
MPEG Results: An Infeasible Set
29/30
0
0.2
0.4
0.6
0.8
1
1.2
0 1 2 3 4 5 6
Time (sec)
Sh
ort
fall
in T
imel
y T
hro
ug
hp
ut
Weighted-delivery
Time-based
Random
DCF
MPEG Results: An Infeasible Set
29/30
0
0.2
0.4
0.6
0.8
1
1.2
0 1 2 3 4 5 6
Time (sec)
Sh
ort
fall
in T
imel
y T
hro
ug
hp
ut
Weighted-delivery
Time-based
Random
DCF
EDCF
Conclusion Extend a framework for QoS to deal with traffic
patterns, deadlines, throughputs, delivery ratios, and channel unreliabilities
Characterize when QoS is feasible
Provide efficient scheduling policies
Address implementation issues
30/30
Backup Slides An example:
Two clients, τ = 3 p1=p2=0.5 q1=0.876, q2=0.45 w1=1.76/3, w2=0.3 I{1}=I{2}=1.25/3, I{1,2}=0.25/3
w1+I{1}=3.01/3 > 1 However, w1+w2+I{1,2}=2.91/3 < 1
top related