21.01.2014 Views

Xiao Liu PhD Thesis.pdf - Faculty of Information and Communication ...

Xiao Liu PhD Thesis.pdf - Faculty of Information and Communication ...

Xiao Liu PhD Thesis.pdf - Faculty of Information and Communication ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

samples from the scientific workflow system logs which are above D ( a i ) or below<br />

d ( a i ) are hence discarded as outliers. The actual runtime duration at a i is denoted<br />

as R ( a i ) . Now, we propose the definition <strong>of</strong> probability based temporal consistency<br />

which is based on the weighted joint distribution <strong>of</strong> activity durations. Note that,<br />

since our temporal framework includes both build-time components <strong>and</strong> runtime<br />

components to support the lifecycle <strong>of</strong> scientific workflow systems. Accordingly,<br />

our probability based temporal consistency model also includes both definitions for<br />

build-time temporal consistency <strong>and</strong> runtime temporal consistency.<br />

Definition 5.2: (Probability based temporal consistency model).<br />

At build-time stage, U (SW ) is said to be:<br />

n<br />

1) Absolute Consistency (AC), if ∑ wi ( µ i + 3σ<br />

i ) < U ( SW ) ;<br />

i=<br />

1<br />

n<br />

2) Absolute Inconsistency (AI), if ∑ wi ( µ i − 3σ<br />

i ) > U ( SW ) ;<br />

i=<br />

1<br />

3) α%<br />

Consistency ( α%<br />

C), if ∑ w ( µ + λσ ) u(<br />

SW ) .<br />

n<br />

i=<br />

1<br />

At runtime stage, at a workflow activity<br />

i i i =<br />

a p ( 1 < p < n ), U (SW ) is said to be:<br />

p<br />

n<br />

1) Absolute Consistency (AC), if ∑ R(<br />

ai ) + ∑ wi<br />

( µ i + 3σ<br />

i ) < U ( SW ) ;<br />

i=<br />

1 j=<br />

p+<br />

1<br />

p<br />

n<br />

2) Absolute Inconsistency (AI), if ∑ R(<br />

ai ) + ∑ wi<br />

( µ i − 3σ<br />

i ) > U ( SW ) ;<br />

i=<br />

1 j=<br />

p+<br />

1<br />

p<br />

n<br />

i i i i =<br />

j=<br />

p+<br />

1<br />

3) α%<br />

Consistency ( α%<br />

C), if ∑ R(<br />

a ) + ∑ w ( µ + λσ ) U ( SW ) .<br />

Here<br />

i=<br />

1<br />

w i st<strong>and</strong>s for the weight <strong>of</strong> activity a i , λ ( − 3 ≤ λ ≤ 3 ) is defined as the<br />

α % confidence percentile with the cumulative normal distribution function <strong>of</strong><br />

−(<br />

x−µ<br />

2<br />

1<br />

i )<br />

µ + λσ<br />

F( µ + λσ ) =<br />

2σ<br />

2<br />

∫ i i<br />

i i<br />

−∞<br />

l i • dx = α%<br />

( 0 < α < 100 ).<br />

σ 2π<br />

Note that the build-time model will mainly be used in Component 1 <strong>and</strong> its<br />

runtime counterpart will mainly be used in Component 2 <strong>and</strong> Component 3 <strong>of</strong> this<br />

thesis. Meanwhile, as explained in [56], without losing generality, the weights <strong>of</strong><br />

78

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

Saved successfully!

Ooh no, something went wrong!