DARPA ULTRALOG Final Report - Industrial and Manufacturing ...
DARPA ULTRALOG Final Report - Industrial and Manufacturing ...
DARPA ULTRALOG Final Report - Industrial and Manufacturing ...
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
Manuscript for IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS 16<br />
w<br />
*<br />
i<br />
=<br />
T<br />
p∈K<br />
*<br />
i<br />
∑<br />
T<br />
n(<br />
i)<br />
*<br />
p<br />
ω<br />
n(<br />
i)<br />
(22)<br />
Component’s decision<br />
v<br />
*<br />
i<br />
*<br />
1 ( Ti<br />
− Ri<br />
( t)<br />
= f<br />
− i ( )<br />
(23)<br />
L ( t)<br />
i<br />
4.2 Analysis<br />
Resource manager’s bidding function b n (T) in (19) can be composed referring to the solution<br />
algorithm of fractional knapsack problem. In the fractional knapsack problem, there are multiple<br />
items that can be broken into fractions. Given unit weight <strong>and</strong> unit value of each item, the<br />
problem is to determine the amount of each item so as to maximize total value subject to a<br />
weight capacity. The fractional knapsack problem can be easily solved by a greedy algorithm,<br />
i.e., take as much as possible of the item that is the most valuable per unit weight until the<br />
capacity is reached. Similarly, b n (T) can be composed using a greedy algorithm. As b i (T i ) in (18)<br />
is a piecewise-linear increasing concave function, take the most valuable piece per unit T i among<br />
the first available pieces until all pieces are taken. This greedy algorithm leads building the<br />
resource manager’s bidding function in O(|K n | 2 ), where |X| denotes the cardinality of set X.<br />
Similarly, resource manager’s decision problem in (21) can be solved in O(|K n | 2 ), using the<br />
greedy algorithm except that (fractional) pieces are taken until a capacity is reached. So, the<br />
complexity of all resource managers’ local problems is O(|A| 2 ) in the worst case when |N|=1.<br />
The seller’s decision problem in (20) is simply a single variable problem, which can be solved<br />
using diverse search methods depending on the structure of objective function. As each b n (T) is<br />
piecewise-linear increasing concave function, ∑b n (T) is also a piecewise-linear increasing<br />
concave function <strong>and</strong> its number of pieces is proportional to |A|. To compose ∑b n (T) from each