Control Theoretical Model for QoS Adaptations

actuallyabandonedElectronics - Devices

Nov 15, 2013 (3 years and 4 months ago)

84 views

Control Theoretical Model for QoS Adaptations


Goal: develop a Task Control Model to formally model
applications that adapt to resource/QoS variations


Use digital control theory


Analyze equilibrium, stability and fairness


Model realistic cases where complete task state
information are not observable


Map to an adaptive control system


-

Target System

to be controlled takes actions to process
input


-

input is determined by a
Controller


-

controller monitors states or output of target system and
compares them to desired values (the
reference
)

Task Flow Model


Consider each application as an ensemble of functional
components (
tasks
)


A task performs certain actions to deliver a result to
other tasks or end
-
user


Task Flow Graph is a directed acyclic graph showing
dependencies among tasks


A task can be uniquely characterized by its input
quality, output quality and utilized resources (needed
to perform actions)

Task Control Model


Models a single task in the Task Flow Graph


This is the
Target Task

to be controlled


In addition,


-

Adaptation Task

performs the adaptive control
algorithm


-

Observation Task

observes states of the Target Task
and feeds them back to Adaptation Task

Task Control Model (cont’d)


Adaptation Task modifies a set of
controllable

parameters,
i.e.

possible to affect their values, which
in turn can affect states of Target Task and thus its
output quality


Task States characterize the internal dynamics in the
Target Task. The most important states are its
parameters related to its resources


Observation Task observes task states if they are
observable
. If not, estimates or predicts the current
states

Control Equations


Task is said to be at
equilibrium/stable

when its state
does not change


Assume Target Task can be characterized accurately
by discrete
-
time equations

Example


Resource requests for temporal or spatial resources


Request rate of a task throttled by Adaptation Task,
so that it does not exceed its fair share


Assume to observe the
total

number of outstanding
resource requests made by
all

tasks at time k


Under PID control, find stability conditions,
equilibrium states, and evaluate responsiveness
configurability


Show stability, equilibrium and configurability are
preserved when Observation Task can only observe
the number of outstanding requests made by the task
itself