Incremental Materialization of RDF Graph Closures for Stream Reasoning Alexandre Mello Ferreira (PhD student) 22/11/2010.
Post on 19-Dec-2015
220 Views
Preview:
Transcript
Incremental Materialization of RDF Graph Closures for Stream Reasoning
Alexandre Mello Ferreira (PhD student)
22/11/2010
Alexandre Mello FerreiraDEI
2Outline
Introduction
Problem statement
RDF in a nutshell
Our proposed approach
Envisioned scenario
Model
Time-stamped streams
Incremental maintenance of materialization for RDF streams
Implementation and first results
Jena
Deductive rules
Results
Alexandre Mello FerreiraDEI
3Outline
Introduction
Problem statement
RDF in a nutshell
Our proposed approach
Envisioned scenario
Model
Time-stamped streams
Incremental maintenance of materialization for RDF streams
Implementation and first results
Jena
Deductive rules
Results
Alexandre Mello FerreiraDEI
4Problem statement
Large use of sensed data
Urban computing
Green computing
Monitoring systems in large environments lead to complex
and cost hungry systems
Invisibility to keep ontology representation in memory due
to often updates
How to make data representation more useful
E.g. (from IBGE):
<Brazil, hasPopulation, 070mi>
<Brazil, hasPopulation, 192mi>
<Brazil, hasPopulation, 260mi>
Alexandre Mello FerreiraDEI
5Problem statement
Large use of sensed data
Urban computing
Green computing
Monitoring systems in large environments lead to complex
and cost hungry systems
Invisibility to keep ontology representation in memory due
to often updates
How to make data representation more useful
E.g. (from IBGE):
<Brazil, hasPopulation, 070mi> [1960]
<Brazil, hasPopulation, 192mi> [2010]
<Brazil, hasPopulation, 260mi> [2050]
Alexandre Mello FerreiraDEI
6RDF in a nutshell
Resource Description Framework (RDF) is a W3C
recommendation for resource description
Basically composed by: <S, P, O>
SUBJECT: something identified (resource)
PREDICATE: property that describes the subject
OBJECT: either the property value or another resource
POLIMI
DEI
45.48
9.23
“DEI”
“Dipartimento di Elettronica e Informazione”
dc:title
edu:hasDept
geo:long
geo:lat
Alexandre Mello FerreiraDEI
7RDF in a nutshell
Web semantics vocabularies
Serialization syntax:
Notion 3 (“N3”)
RDF/XML
Alexandre Mello FerreiraDEI
8Outline
Introduction
Problem statement
RDF in a nutshell
Our proposed approach
Envisioned scenario
Model
Time-stamped streams
Incremental maintenance of materialization for RDF streams
Implementation and first results
Jena
Deductive rules
Results
Alexandre Mello FerreiraDEI
9Envisioned scenario
Typical 5,000 square-foot data center
Demand side – IT systems
Supply side – Cooling systems and power systems
Power consumption
(watts)
Usage percentage
(%)
Alexandre Mello FerreiraDEI
10Envisioned scenario
Volume
Mid-range
High-end
Server ID
Rack
CPUs (8x)
Diks (4x)
Mode
Virtualization
eligible
AMD
Intel
Server ID
Usage
Consumption
Mode
Server ID
Usage
Consumption
Mode
RESOURCE TYPE STATIC DATA DEDUCTED DATA
Alexandre Mello FerreiraDEI
11Envisioned scenario
RDF/XML sample of the background knowledge
Alexandre Mello FerreiraDEI
12Outline
Introduction
Problem statement
RDF in a nutshell
Our proposed approach
Envisioned scenario
Model
Time-stamped streams
Incremental maintenance of materialization for RDF streams
Implementation and first results
Jena
Deductive rules
Results
Alexandre Mello FerreiraDEI
13Time-stamped streams
Static data
Stream data
Derived data
Time-stamped data
Alexandre Mello FerreiraDEI
14Incremental maintenance
Based on the work developed by Volz and Prof. Ceri
database research group, the following triples definitions
are considered:
Tin enter in the window stream
Tstay stay in the window stream and derived
Texp exit the window (expire) stream and derived
dTnew Triples trigged by Tin and not in Tstay
dTrenew Triples trigged by Tin and in Tstay
dTtimestamp Triples trigged by both dTnew and dTrenew
T+ add to the materialization
T- remove fom the materializationTresult = (Tinicial U T+) \ T-
Alexandre Mello FerreiraDEI
15Incremental maintenance
The implemented solution (time 6):
s1
c592
51
c370
1
2
d2
isConsuming
isConsuming
isConsuming
hasDisk
hasCPU
hasCPU
LowPowerhasMode
47
3isUsing
Alexandre Mello FerreiraDEI
16Incremental maintenance
The implemented solution (time 7):
s1
c554
51
c3
7
2
d2
isConsuming
isConsuming
hasDisk
hasCPU
hasCPU
LowPowerhasMode
LowPower
hasMode
47
3isUsing
Alexandre Mello FerreiraDEI
17Incremental maintenance
The implemented solution (time 7):
s1
c554
51
c3
7
2
d2
isConsuming
isConsuming
hasDisk
hasCPU
hasCPU
LowPowerhasMode
LowPower
hasMode
47
3isUsing
Alexandre Mello FerreiraDEI
18Incremental maintenance
The implemented solution (time 8):
s1
c554
c3
7d2
47
3isConsuming
isUsing
hasDisk
hasCPU
hasCPU
LowPower
hasMode
c9
8
12isUsing
52
hasCPU
isConsuming
LowPower
hasMode
quietMode
hasMode
True
eliVirt
Alexandre Mello FerreiraDEI
19Outline
Introduction
Problem statement
RDF in a nutshell
Our proposed approach
Envisioned scenario
Model
Time-stamped streams
Incremental maintenance of materialization for RDF streams
Implementation and first results
Jena
Deductive rules
Results
Alexandre Mello FerreiraDEI
20Jena2 inference subsystem
Framework to develop semantic web app
It provides:
RDF API
Reading and writing RDF/XML, N3, and N-Triples
In-memory and persistence storage
SPARQL query engine
1. // creates an empty RDF model2. Model myRDFmodel = ModelFactory.createDefaultModel();3. 4. // creates a new generic rule reasoner to support user defined rules5. Reasoner reasoner = new GenericRuleReasoner(Rule.parseRules(ruleSrc));6. reasoner.setDerivationLogging(true);7. 8. // creates a new inference model which performs RDF inference over myRDFmodel 9. // using my previous defined reasoner10. InfModel inf = ModelFactory.createInfModel(reasoner, myRDFmodel);
Alexandre Mello FerreiraDEI
21Deductive rules
Alexandre Mello FerreiraDEI
22Deductive rules
Alexandre Mello FerreiraDEI
23Deductive rules
Alexandre Mello FerreiraDEI
24First results
0 5 10 15 20 251
10
100
1000
10000
naive approach incremental-stream
ms.
Average time to maintain the materialization vs window sliding
Axis X represents the number of arrival stream triples
It depends on the type of the triple
Alexandre Mello FerreiraDEI
25First results
0 100 200 300 400 500 600 700 800 900 10001
10
100
1000
10000
naive approach incremental-stream
ms.
Average time to maintain the materialization vs incremental
number of monitored data (sensors)
Axis X represents the number of sensed components
It keeps homogeneous regarding to scalability
Alexandre Mello FerreiraDEI
26Conclusion remarks
Next steps
Merge our scenario with Urban computing in order to come up
with comparable experiments
Try alternative inference engines and compare their features
Apply the proposed approach to a real data center environment
(like in GAMES project)
Alexandre Mello FerreiraDEI
27
Thank you!
top related