Transcript presentation

Recording the Context
of Action for Process
Documentation
Ian Wootten
Cardiff University, UK
[email protected]
Context

Definitions:
 Circumstances forming the setting for an event, statement or idea
[Oxford English Dictionary, 2008]



User environment elements a computer knows about [Brown, 1995]
Characterisation of the situation of entities [Dey and Abowd, 2000]
Properties which can support/dispute evidence of actions
 More than component interaction
 More informed judgements can be made
 Subjective in nature
 Ad-hoc documentation between applications
 Records of data with unknown relationships could be useful
 May help out at a later date
Process Distinction

Provenance is about processes

“The process which led to some data”




[Groth et al. 2006]
Sequences of actions
How did this come to be the way it is?
Achieved by:
 Documenting relationships, component interaction
 Evidence
If actions in a process are the same, locating distinct
traces becomes more difficult

e.g. I invoke this workflow multiple times, are any records
unique? Were they performed in different situations?
Context Uses

Automatic assertion in legacy actors


Prediction of future actor properties



Record context and actions
Probabilistic model constructed
Similarity of past process traces


E.g. Long running, data mining services
Context recorded and compared for two
provenance traces
And others….
Documenting Process

Cannot answer all provenance queries
with documentation of interaction alone

Eg. What was happening to cause such
behaviour? Why does execution of the
same workflow result in different execution
times? How do we know an action is
subject to the same conditions?
Host
Invoke
arg
Actor
f1()
f1(arg)

We know nothing of the context under
which assertions are made


Answers can be given by entities
themselves (e.g using PReServ)
Particular focus on deriving context from
measurable values
f2()
Result
Result
f2(f1(arg))
Time Series Knowledge
Representation (TSKR)
Properties and States for an
actor are represented using
the TSKR [Moerchen 2006]
Series extracted from several
numerical variables

S5
S6
S2
S7
S1
Coincidence intervals found
B
A
S2
S3
S4
F
D
AD
C
D
C
F
B
F
BE
BD
AF
S5
S6
S2
11
BD
D
31
Patterns
AD
E
12
States represented in transition
table
Chords
/States
F
22
D
Monitored variables specified
by service administrator
S1
C
Tones
32

B
S4
Segmentation, Shape-based
Resultant series shows time
intervals when multiple
conditions occur (states)

A
S3
33

S2
13

S1
12

11

S7
S1
Documenting Context

Provide a mechanism to specify and
automatically record environmental
context for any application



Our policy configuration

Gathers monitoring data and mines
states using TSKR
M
Registry
Atlas
Image
Atlas
Header
Observer

Monitoring
Policy
Observer

Capture using process documentation as
assertions of actor state, using PReServ
Operate according to a particular owner
defined policy
Triggered on service execution through
service wrappers
Reuse existing monitoring resources
(Ganglia, Nagios) through plug-ins
PAssertion
Host System
Slicer
Atlas
Slice
Client
I1
StAR
I2
I3
Monitoring
Sources
Experimentation

Ran two services from provenance challenge 1000 times


Context recorded as actor state assertions
Action recorded as interaction assertions

TSKR Transition history
mapped to a transition table
Used as a predictive tool

TSKR series patterns can be used
for comparison of states



Where series is segmented
Where vast collections of data
need to be explored
Based upon


Context component distances
Maximum distance possible
Pattern Elements

State q
State r
1
1
1
2
1
3
Prediction Results

Three approaches:



State prediction
(TSKR)
Random (with
history)
Random
Similarity Results

Indicates small subsets of documentation
Conclusions

Context helps to understand evidence



For processes realised using SOA, understanding records
of action
Actions may be the same but performed in different
circumstances
TSKR is a good fit for context measurement



Registry approach assists in context capture
Automates the collection of actor state
Demonstrated as:
 Good predictor of state
 Useful identification of state properties