01.12.2012 Views

Architecture of Computing Systems (Lecture Notes in Computer ...

Architecture of Computing Systems (Lecture Notes in Computer ...

Architecture of Computing Systems (Lecture Notes in Computer ...

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

42 P. Bellasi, W. Fornaciari, and D. Siorpaes<br />

“Feasible System-wide Configurations” (FSC) available for the target system,<br />

that we named: FSC model. A FSC is a region on the SWCS, thus is def<strong>in</strong>ed<br />

by a set <strong>of</strong> validity ranges for each SWM, where it is granted that each device<br />

could be configured to operate <strong>in</strong> a WM that does not have any conflict with any<br />

other device. These regions are particularly important because they grant that<br />

any <strong>in</strong>ter-dependency among devices is safely solved. Though a number <strong>of</strong> <strong>in</strong>terest<strong>in</strong>g<br />

theoretical techniques can be def<strong>in</strong>ed to identify at run-time what is the<br />

optimal system configuration, accord<strong>in</strong>g to both the available resources and the<br />

required performance, every outcome is useless if it cannot be actually applied to<br />

the real system because <strong>of</strong> implicit <strong>in</strong>ter-dependencies or hardware constra<strong>in</strong>ts<br />

ignored by the optimization policy itself. Indeed an optimized configuration cannot<br />

be identified regardless <strong>of</strong> its feasibility. Thanks to their <strong>in</strong>terest<strong>in</strong>g property,<br />

the identification <strong>of</strong> all system’s FSCs is especially important. Consider<strong>in</strong>g this,<br />

the optimization technique proposed is based on the ’a-priori identification’ <strong>of</strong><br />

all and only the system feasible configurations. Thus, any optimization policy<br />

that will be developed on top <strong>of</strong> this framework, it will be granted to operate on<br />

a set <strong>of</strong> real and valid configurations and consequently each result can be safely<br />

applied to real system.<br />

The Optimization Layer exploits the system view <strong>of</strong>fered by the underly<strong>in</strong>g<br />

FSC model to support the global optimization policy <strong>of</strong> the proposed hierarchical<br />

distributed control. This is obta<strong>in</strong>ed by the def<strong>in</strong>ition <strong>of</strong> a strategy to assign<br />

a “weight” to each feasible configuration accord<strong>in</strong>g to the runn<strong>in</strong>g optimization<br />

policy. The weight associated to FSC is def<strong>in</strong>ed to be a sufficiently abstract metric<br />

which can be easily adapted for a generic multi-objective optimization. The<br />

run-time track<strong>in</strong>g <strong>of</strong> application requirements is another goal <strong>of</strong> this layer. The<br />

abstract system model, based on the concept <strong>of</strong> FSC and their representation<br />

<strong>in</strong> the SWCS, is properly exploited at this layer to translate application requirements<br />

on constra<strong>in</strong>ts for the research <strong>of</strong> the optimal configuration. Indeed,<br />

application requirements are translated on constra<strong>in</strong>ts for the optimization problem<br />

which could <strong>in</strong>validates some FSCs. Thus, this layer provides support for<br />

both: FSC pre-order<strong>in</strong>g, accord<strong>in</strong>g to the optimization objectives <strong>of</strong> the runn<strong>in</strong>g<br />

policy, and optimal FSC selection, consider<strong>in</strong>g user-space requirements to filter<br />

out run-time <strong>in</strong>validated FSC.<br />

3.3 Formal Validation <strong>of</strong> the Optimization Policy<br />

The hierarchical distributed control problem can be conveniently reformulated<br />

us<strong>in</strong>g an appropriate formal model. A transposition <strong>of</strong> this type has been done<br />

not only to provides a rigorous description <strong>of</strong> the problem and a formal pro<strong>of</strong><br />

<strong>of</strong> the solution quality, but also allows to identify more easily a possible alternative<br />

solution strategy by tak<strong>in</strong>g <strong>in</strong>to account the particularities <strong>of</strong> the specific<br />

formulation. We have reformulated the problem us<strong>in</strong>g L<strong>in</strong>ear Programm<strong>in</strong>g.<br />

A simple representation <strong>of</strong> the LP formulation is depicted <strong>in</strong> Fig. 3. In this<br />

simple scenario we consider a system with three devices (d1, d2 and d3) and<br />

two SWMs (p1 and p2). The available FSCs are only three, but at run-time<br />

it could happen that some <strong>of</strong> them (FSC3 <strong>in</strong> the example) are <strong>in</strong>validated by

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!