Admission Control and Sched uling for QoS Guarantees fo r Variable-Bit-Rate Applica tions on Wireless Channels I-H. Hou and P.R. Kumar Department of Computer Science University of Illin ois 報報報 報報報 : Proc. of ACM MobiHoc, pages 175– 184, 2009
Admission Control and Scheduling for QoS Guarantees for Variable-Bit-Rate App
lications on Wireless Channels
I-H. Hou and P.R. Kumar
Department of Computer Science University of Illinois
報告人:李宗穎
Proc. of ACM MobiHoc, pages 175–184, 2009
2
Outline
Introduction A model for QoS with Probabilistic Arrival Example for Application Necessary Condition for Feasibility Scheduling Policies Implementation and Simulation
3
Introduction
Two major issues concerning QoS over wireless admission control scheduling
describe the QoS requirements by four criteria traffic pattern channel reliability delay bound throughput bound
4
A model for QoS with Probabilistic Arrival (1/7)
QoS requirements that generalizes a model that has been proposed in [5]
[5] I-H. Hou, V. Borkar, and P. R. Kumar. A theory of QoS for wireless. To appear in Proc. of INFOCOM 2009.
5
A model for QoS with Probabilistic Arrival (2/7)
Clients generate jobs for the server to accomplish, and during each time slot, the server can attempt exactly one job
The time slots are grouped into intervals, with each interval containing τ time slots
Unfinished jobs are discarded at the end of an interval, so a delay bound of τ time slots is imposed on all jobs
6
A model for QoS with Probabilistic Arrival (3/7)
Don’t restrict attention only to clients that generate one job during each interval
Clients generate jobs according to a probability mass function and exactly every client in S generates a job in an interval is R(S)
7
A model for QoS with Probabilistic Arrival (4/7)
Because wireless channels are unreliable, the job gets delivered with probability pn, which is called the reliability for client n
Each client n requires a long-term average throughput of qn delivered jobs per interval
8
A model for QoS with Probabilistic Arrival (5/7)
Definition 1.
Let Ht be the set of all possible histories of the system up to time slot t. A scheduling policy is a function η: Ht{1, 2,…, N, ψ} with the interpretation
at time slot t + 1, the server attempts to transmit the job from client n if η(ht) = n or idles if η(ht) = ψ
9
A model for QoS with Probabilistic Arrival (6/7)
Definition 2. A set of clients is said to be fulfilled by a sc
heduling policy η if the long-term average throughput of each client n is at least qn jobs per interval with probability 1
qn the timely throughput bound of client n
10
A model for QoS with Probabilistic Arrival (7/7)
Definition 3. A set of clients is said to be feasible if there
exists a scheduling policy η that fulfills it
Definition 4. An optimal scheduling policy is a policy
that fulfills every feasible set of clients
11
Video Stream
MPEG alternates between three coding modes (I,P,B) that require different numbers of bits per frame
a higher bit rate implies a higher arrival probability, and can be converted into a timely throughput bound, and captured by the parameter qn
12
VoIP Stream
VoIP traffic involves both uplink traffic and downlink traffic
This paper consider audio codecs that generate CBR traffic, such as ITU-T G.711
The job generation time for clients may be offset a set of three clients {1, 2, 3} client {1 (1,3,5…) 2 (2,4,6…) 3 (1,4,7…)} R(1, 3) = R(2, 3) = 1/2
13
Real Time Surveillance
there may be sensor nodes for monitoring heart activity, blood pressure, and body temperature
Paper assume each client generates jobs periodically, with the differing frequencies of job generation reflecting the importance of the corresponding data
14
Necessary Condition for Feasibility (1/6)
Paper extend some Lemma to a necessary condition in [5] for a set of clients to be feasible to the more general model with variable traffic arrival patterns
[5] I-H. Hou, V. Borkar, and P. R. Kumar. A theory of QoS for wireless. To appear in Proc. of INFOCOM 2009.
15
Necessary Condition for Feasibility (2/6)
LEMMA 1. The long-term average timely throughput of
a client n is at least qn jobs per interval if and only if the server, on average, attempts jobs from that client wn = qn/pn times per interval
16
Necessary Condition for Feasibility (3/6)
LEMMA 2. A set of N clients is feasible only if ΣN
wn≦ τ Since the length of an interval is τ time slots
and the server can attempt jobs at most once in each time slot
17
Necessary Condition for Feasibility (4/6)
LEMMA 3. γn be the random variable denoting the number
of attempts the server needs to make for a job from client n Prob{γn = t} = pn(1 - pn)t-1 (Geometric dist.)
,,0
,,,
otherwise
ifL Sn nSn n
S
Delay Bound Guarantee ?
18
Necessary Condition for Feasibility (5/6)
LEMMA 4. E[LS]: expected number of idle time slots in interval R(S): an interval occurs with probability
N
n Ssn LESRw
1
][)(
S
SSS LESRI ][)( ''
19
Necessary Condition for Feasibility (6/6)
LEMMA 5. A set of clients is feasible only if hol
ds for every subset S It may seem that the condition for a strict subset
S of {1,2,…,N} is redundant, and that we only need to evaluate the condition for all clients
Sn Sn Iw
20
Example
interval length τ = 3, and two clients Client 1 p1=0.5 q1=0.876 R{1}=1
Client 2 p2=0.5 q2=0.45 R{2}=1
So We can calculate following value w1 = 1.76 ; w2 = 0.9
I{1} = I{2} = 1.25 ; I{1,2} = 0.25
w1 = 1.76 > 1.75 = τ – I{1} (unfeasible!!) w1 + w2 = 2.66 > 2.75 = τ – I{1,2} (feasible)
21
largest time-based debt first policy
DEFINITION 6. Let un(t) denote the number of attempts that
the server has made for jobs from client n up to time slot t. The time-based debt for client n is defined to be wnt - un(t)
22
largest weighted-delivery debt first policy
DEFINITION 7. Let cn(t) denote the number of jobs for clien
t n accomplished by the server up to time slot t. The weighted delivery debt for client n is defined to be [qnt - cn(t)]/pn
24
VoIP traffic
Group A : 60ms, 21.3kbits/s, 99% delivery ratio Subgroup {A1, A2, A3}, Ai begin i, i+3, i+6…
Group B : 40ms, 32kbits/s, 80% delivery ratio Subgroup {B1, B2}, Bj begin j, j+2…
Feasible set 6 clients in each subgroup Ai,
5 clients in each subgroup Bj (infeasible Bj=6) The channel reliability of the nth client in each subgroup
is (60 + n)%
26
MPEG Video Streaming
Group A:0.765 packet/interval, 90% delivery ratio Group B:0.34 packet/interval, 80% delivery ratio Feasible set 4 clients in each subgroup Ai,
4 clients in each subgroup Bj (infeasible Aj=5) The channel reliability of the nth client in each subgroup
is (60 + n)%