Download as pdf or txt
Download as pdf or txt
You are on page 1of 2

Solving Joint Network and Computing Resource Scheduling

Problem by Recursive Solution Partition and Elimination

I. I NTRODUCTION Segment order constraints:


In this paper the authors propose a novel strategy to solve fi,b ≥ fi,a,Ci,a + 1 − M ∗ ui
JRS problem using solution enumeration and elimination un-
fi,a,l+1 ≥ fi,a,l + 1 − M ∗ ui , l ∈ {1, 2, ..., Ci,a − 1}
der an integer linear programming formulation. The interest
(3)
of this method does not lie in the detailed techniques, but in a
higher level strategy of reformulation. The major contribution Network resource constraints:
of this paper is the methodology of using solution space
partition and elimination, which not only can be applied to fi,a,p 6= fj,a,q − 1 ∀i 6= j (4)
JRS problem but also to other similar applications.
where p ∈ {1, 2, ..., Ci,a }, q ∈ {1, 2, ..., Cj,a }.
II. TASK M ODEL AND P ROBLEM F ORMULATION CPU resource constraints:
A. Task Model
fi,b 6= fj,b − 1 ∀i 6= j (5)
In our JRS model, each real-time task τ : τa (x) → τb (m)
consists of two consecutive and dependent segments: network
segment τa , computing segment τb . We use x, m to represent
the execution time of the three segments, respectively. More C. Algorithm Overview
specifically, the execution time of the sensing and actuating
segments are their corresponding network transmission time. The insight of this algorithm is that if the computing
In this paper, we assume that the unit size of network and segments are fixed for all tasks, their corresponding network
computing resource slices are the same. The wireless network segments are easy to be scheduled.
has only one channel and the CPU is a uniprocessor. These Theorem 1. Add a theorem here
assumptions will be further relaxed in the future work.
Let the sub-problem(SP) to be the scheduling of computing
B. Problem Formulation segments only. Given a solution F̂ of SP, F̂ can be viewed
Joint Resource Scheduling (JRS) Problem: Given a set as part of a solution F of the original problem. While there
of real-time tasks {τ1 , τ2 , ..., τn }. Each task τi is associated are multiple F containing F̂ , the optima can be found in
with a release time ri , a deadline di , and two segments τi,a , polynomial time. Thus, we take advantage of this relationship
τi,b . Ci,a and Ci,b denote the execution time involved in the and perform branch and bound on the solution space of SP
network segment and the computing segment. In this paper instead of MP. Anyone who familiar with Dantzig Wolfe
we assume τi,b must be executed continuously while τi,a can Decomposition and Benders’ Decomposition will find it easy
be separated into several intervals. Though the assumption is to understand the strategy, however, the authors do not view
specific, we claim that our algorithm can be applied to more this algorithm as an application, nor modification of any
general scenarios with minor modification. The objective of existing algorithms, because the solution space partition and
the JRS problem is to find a schedule minimizing the number elimination, as will be descried bellow is essential.
of dropped tasks.
D. Algorithm Description
We first formulate the JRS problem as an integer linear
programming (ILP) problem. It aims to construct the schedule Let the original JRS problem be M. Let F̂ be one feasible
Snet = {fi,a,k } for the network segments and Scom = {fi,b } schedule of the a set of computing segments F̂ = {fi,b }, and
for the computing segments, where fi,a,k , fi,b are the finish denote M(F̂ ) the sub problem seeking for a feasible schedule
time of the k th unit of segments τi,a and τi,b , respectively. Let with the fixed computing segments. According to the theorem,
ui be the indicator if task τi is dropped. Formally, M(F̂ ) can be solved in polynomial time. Thus the original M
Objective: can be reformulated to find the scheduling F̂ that minimizes
X the original objective. Next we try to enumerate all the feasible
Obj = Min (ui ) (1)
i
F̂ .
Let the solution space of F̂ be F, for small instances, we
Release time and deadline constraints: can count every single feasible solution in F and assign each
fi,a,1 ≥ ri + 1 − M ∗ ui of them a binary label xi . Next we reformulate M in the
(2) following way:
fi,b ≤ di + M ∗ ui
Objective: X
Obj = Min (ui ) (6)
i

Release time and deadline constraints:

fi,a,1 ≥ ri + 1 − M ∗ ui (7)

Segment order constraints:


fi,a,l+1 ≥ fi,a,l + 1 − M ∗ ui , l ∈ {1, 2, ..., Ci,a − 1}
(8)
Network resource constraints:
fi,a,p 6= fj,a,q − 1 ∀i 6= j (9)
where p ∈ {1, 2, ..., Ci,a }, q ∈ {1, 2, ..., Cj,a }.
Computing segment scheduling constraints:
X
xi = 1 (10)
i

Relationship constraints:
X
fi,a,Ci,a ≤ xj ż (11)
j

where fi,b = z in xj .
III. C ONCLUSIONS AND F UTURE W ORK
In this paper, we report our progress on the study of the
joint network and computing resource scheduling problem in
WNCSs.
R EFERENCES

You might also like