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.
denotes the immediate predecessors of component i. Then, by denoting CPU time per task in the<br />
given topology as P i , LI i is represented as in (5).<br />
∑<br />
L i = rti<br />
+ La<br />
(4)<br />
i<br />
a∈i<br />
LI = L P . (5)<br />
To provide theoretical foundation of optimal resource allocation, we convert a network into a<br />
network with infinitesimal tasks. Each root task is divided into r infinitesimal tasks <strong>and</strong> each P i is<br />
replaced with P i /r. Then, the load index of each component is the same as the original network<br />
but tasks are infinitesimal. We denote the completion time of the network with infinitesimal<br />
tasks as T´. Also, we define a term called task availability as an indicator of relative preference<br />
for task arrival patterns. A component’s task availability for an arrival pattern is higher than for<br />
another if cumulative number of arrived tasks is larger or equal over time. A component prefers a<br />
task arrival pattern with higher task availability as it can utilize more resource. Consider a<br />
network <strong>and</strong> reconfigure it such that all components have their tasks in their queues at t=0. Each<br />
component has maximal task availability in the reconfigured network <strong>and</strong> the completion time of<br />
the reconfigured network forms the lower bound T LB of a network’s completion time T given by:<br />
i<br />
i<br />
T<br />
LB<br />
ωk<br />
= Max ∑ LI i<br />
k∈K<br />
a . (6)<br />
ω<br />
k i∈<br />
S I<br />
[ k ]<br />
Then, assuming a hypothetical weighted round-robin server 2 for CPU scheduling, T´ equals to<br />
T LB when each machine allocates resource to the residing components according to (7), where<br />
k(i) denotes a machine in which component i resides.<br />
2 The hypothetical server has idealized fairness as the CPU time received by each thread in a round is infinitesimal<br />
<strong>and</strong> proportional to the weight of the thread. This assumption is reasonable because quantum size is relatively<br />
infinitesimal compared to working horizon in reality.<br />
8