Optimality of Periodwise Static Priority Policies in Real-Time Communications
I-Hong Hou, Anh Truong, Santanu Chakraborty, P.R. Kumar
1
Motivation Study the scheduling policies for real-time
wireless communication
Each packet has a strict deadline Timely-throughput: the throughput of packets
that are delivered on time Consider the unreliable nature of wireless
transmissions
Previous work has proposed scheduling policies
This work: understand some properties of the policies
2
Client-Server Model
3
A system with N wireless clients and one AP Time is slotted AP schedules all transmissions
AP
1
2
3
Traffic Model
4
Each client generates one packet every T time slots T time slots form an period
AP
1
2
3
T
Delay Bounds
5
Deadline for each packet = T Packets are dropped if not delivered by the
deadline Delay of successfully delivered packet is at
most T
AP
1
2
3
T
Channel Model
6
Transmissions are unreliable A transmission to client n succeeds with
probability pn
AP
1
2
3
T
p1
p2
p3
A Scheduling Example
7
AP
1
2
3
p1
p2
p3
F S
S
F F
Packet expires and is dropped
S
S
S I I
I I
I I
Forced idleness
Timely Throughput
Timely Throughput = long-term average # of packets received in a period
AP
1
2
3
8
p1
p2
p3
F S
S
F F
S
S
S I I
I I
I I
Timely Throughput
0.5
1.0
1.0
Timely Throughput Requirements
Client n requires timely throughput qn System is fulfilled if all requirements are
met
AP
1
2
3
9
p1
p2
p3
F S
S
F F
S
S
S I I
I I
I I
Timely Throughput
0.5
1.0
1.0
qn
0.7
0.3
0.5
Summary of the Model
Clients have strict per-packet delay bound Clients have timely throughput
requirements Wireless transmissions are unreliable
AP
1
2
3
10
p1
p2
p3
F S
S
F F
S
S
S I I
I I
I I
Largest Debt First Policy
Give higher priority to client with larger “debt”
AP
1
2
3
11
p1
p2
p3
Largest Debt First Policy
Give higher priority to client with larger “debt”
AP
1
2
3
12
p1
p2
p3
F
F
S
F S
Optimality Result Theorem: By choosing the right definition
of debt, the largest debt first policy fulfills all feasible systems Adapt debt according to (qn - actual timely
throughput)
Therefore, it is a Feasibility Optimal Policy
The AP does not need to change ordering during the period
Q: Why the AP doesn’t need to change ordering?
13
How many time slots per period does client n need to obtain a timely throughput of qn?
Ans: There are times that the AP is forced to be idle Let IS = Expected number of idle time slots
when the set of clients is S Theorem: A system is feasible if and only if
Feasibility Constraints
14
∑𝑛∈𝑆
𝑞𝑛
𝑝𝑛≤𝑇 − 𝐼𝑆=: 𝑓 (𝑆)
Time we need to work on S
Time we can work on S
How many time slots per interval does client n need to obtain a timely throughput of qn?
Ans: There are times that the AP is forced to be idle Let IS = Expected number of idle time slots
when the set of clients is S Theorem: A system is feasible if and only if
Feasible region: The region consists of all feasible [qn]
Feasibility Constraints
15
∑𝑛∈𝑆
𝑞𝑛
𝑝𝑛≤𝑇 − 𝐼𝑆=: 𝑓 (𝑆)
Flow of Arguments
16
Periodwise Priority policy can be feasibility optimalVertices of the feasible region can be achieved by some priority ordering among clients
Feasible region forms a polymatroid
f(S) (= T – IS ) is submodular
Any feasible [qn] is a convex combination of vertices of the feasible region
Hence, it can be achieved by time-sharing among priority orderings corresponding to the vertices
17
Periodwise Priority policy can be feasibility optimalVertices of the feasible region can be achieved by some priority ordering among clients
By [D. D. Yao, 2002]
18
Vertices of the feasible region can be achieved by some priority ordering among clients
Feasible region forms a polymatroid
Definition of polymatroid: 1. 2. is non-decreasing 3. is submodular
19
Feasible region forms a polymatroid
f(S) (= T – IS ) is submodular
Let be the expected amount of time that the AP spends on a subset A, if the AP schedules clients in A right after all packets for clients in subset B are delivered
Clearly, is non-increasing with
We can establish that is sub-modular by using this property
Therefore, there exist a periodwise priority policy that is feasibility optimal
20
f(S) (= T – IS ) is submodular
Extension for Time-Varying Channels Wireless channels are time-varying In the period, the channel reliability for client
n is Joint Debt-Channel Policy: Prioritize clients by (debt) [Hou and Kumar 10] has only shown that this policy is feasibility optimal among all periodwise priority policies Now, we can show that this policy if feasibility optimal among all policies
21
AP
1
2
3
p1(t)
p2(t)
p3(t)
Conclusion Study the scheduling policy for real-time
wireless communication
Understand why that a periodwise priority policy can be feasibility optimal
It is because that the feasibility constraints form a polymatroid
Our result can be extended to time-varying wireless channels, and hence establish a previous policy is indeed feasibility optimal22