=Paper=
{{Paper
|id=Vol-1442/paper_33
|storemode=property
|title=Measurement Ontology Pattern Language Applied to Networking Performance Measurement
|pdfUrl=https://ceur-ws.org/Vol-1442/paper_33.pdf
|volume=Vol-1442
|dblpUrl=https://dblp.org/rec/conf/ontobras/NunesVCA15
}}
==Measurement Ontology Pattern Language Applied to Networking Performance Measurement==
https://ceur-ws.org/Vol-1442/paper_33.pdf
Measurement Ontology Pattern Language Applied to
Network Performance Measurement
Raphaela P. Nunes1, Adriana S. Vivacqua1,
Maria Luiza M. Campos1, Ana Carolina Almeida2
1
Programa de Pós-Graduação em Informática - Universidade Federal do Rio de
Janeiro (UFRJ)
Caixa Postal 68.530 – 21941-590 – Rio de Janeiro – RJ - Brasil
2
Instituto de Matemática e Estatística - Departamento de Informática e Ciências da
Computação - Universidade do Estado do Rio de Janeiro (UERJ)
Rua São Francisco Xavier, 524 - 6º andar - 20550-013 - Rio de Janeiro - RJ - Brazil
{mluiza,raphaela.nunes}@ppgi.ufrj.br, avivacqua@dcc.ufrj.br,
ana.almeida@ime.uerj.br
Abstract. The Measurement Ontology Pattern Language (M-OPL) addresses
the measurement core conceptualization according to an Ontology Pattern
Language (OPL). An OPL provides holistic support for solving ontology
development problems for a particular field and guiding the development of
ontologies. This paper presents the application of M-OPL in a specific
domain, network performance measurement. As a result of this application, a
discussion of the use of M-OPL is presented together with some suggestions of
extensions to contemplate the peculiarities of this domain.
1. Introduction
Measurement is a very important discipline in several domains, since it provides useful
information for getting conclusions and making decisions [BARCELLOS et al., 2014].
Measurement is the process of assigning numbers or symbols to properties of real-world
entities, according to widely-defined rules, in order to describe them [FINKELSTEIN;
LEANING, 1984]. It can also be understood as a process that involves a set of actions in
order to characterize entities assigning values to their properties [BARCELLOS et al.,
2014].
When analyzing different areas where measurements can be applied, it is
possible to identify some particular concepts related to the knowledge treated on each
specific area. However, it is also possible to identify some core concepts that are
independent of the application domain. In order to homogeneously represent these core
concepts across different domains, avoiding inconsistencies and ambiguities, it is
important to use a common terminology shared by the domains. Currently, core
ontologies have been used to promote this common conceptualization.
The Measurement Ontology Pattern Language (M-OPL) [BARCELLOS et al.,
2014] addresses the main conceptualization associated to measurements in general,
organized according to an Ontology Pattern Language (OPL). An OPL [FALBO et al.,
2013] corresponds to a network of interconnected ontology modeling patterns that
provides holistic support for solving ontology development problems for a particular
field.
In this work, we describe the application of M-OPL to the scenario of
measurements associated to performance monitoring of Internet links. The objective is
to discuss the scope and usage of M-OPL to generate a new version of the original
ontology developed in the context of the Pinger-LOD Project [SOUZA et al., 2014]. By
aligning it to the core modeling patterns proposed for measurements, we aim to decrease
the possibility of inconsistencies and ambiguities on the ontology and facilitate future
publication and linkage of data with other related data sources in the Web.
During the application of M-OPL we are also concerned to represent
multidimensional aspects of measures, to support the representation of different
perspectives of a measurement. As in other domains, a proposed extension to an existing
OPL may derive, in the future, new modeling patterns to be incorporated in a new
version of the OPL, in this case, an extended version of M-OPL.
The remainder of this paper is organized as follows: Section 2 presents a brief
description of M-OPL. In Section 3, we describe the application process of M-OPL to
the network measurement domain, making firstly a brief explanation of the current
structure of PingER ontology and how it was derived. In Section 4, related works and
further discussions are presented. Finally, in Section 5, we conclude and list some future
work.
2. Measurement Ontology Pattern Language (M-OPL)
A core ontology provides a precise definition of the structural knowledge in a specific
field that spans across several application domains in that field [SCHERP et al., 2012].
Ontology Pattern Languages (OPL) have been proposed and used to organize core
ontologies facilitating their reuse and extension [FALBO et al., 2013]. An OPL
[FALBO et al., 2013] provides a set of interconnected ontology modeling patterns and
a process that describes how to combine them to build an ontology applied to a specific
domain.
The M-OPL addresses the core conceptualization for measurements and their
characterization. M-OPL includes six patterns, defined according to the Unified
Foundational Ontology (UFO) [GUIZZARDI, 2005] and covering six measurements
aspects: Measurement Entities, which include patterns related to the entities and their
properties that can be measured; Measures, which deal with the definition of measures
and classify them according to their dependence on other measures; Measurement Units
& Scales, which concerns the scales related to measures and the measurements units
used to partition the scales; Measurement Procedures, which deals with procedures
required to collect data for measures; Measurement Planning, which addresses the goals
that drive measurement and the measures used to verify goals achievement; and
Measurement & Analysis, which concerns data collection and analysis.
In the application of M-OPL discussed in this article, we will be particularly
interested in applying the Measures aspects, in order to define the measures associated
to the network links measurement domain and characterize the main entity to be
measured (Measurable Entity) in the field, which, in this case, is the Internet Link. For
the purpose of this paper, we are not including details on measurement procedures or
data collection.
3. M-OPL Application to Network Performance Measurement
In this section, we discuss how M-OPL was used to derive a new version of an ontology
for the conceptualization of measurements for network links performance. The original
ontology was developed in the context of the PingER (Ping End-to-end Reporting)
project, which is conducted by the Network and Telecommunications Department at the
SLAC1 National Accelerator Laboratory, in Stanford University, USA. The project
manages data about the quality of Internet links from 1998 to the present day, on an
hourly and daily basis, comprising 16 different metrics collected by 80 monitor nodes
over 800 monitored nodes (more than 8000 pairs of nodes), in more than 160 countries
[COTTRELL, 2001]. Each measurement is basically defined by a ping sent from a
monitor node to a monitored node at any given time, and related to a specific network
metric, considering data packets sizes of 100 and 1000 bytes.
3.1. Original PingER Ontology
The original PingER ontology [SOUZA et al., 2014] was developed to serve as a
reference vocabulary and structure to represent and annotate PingER data as RDF triples
for a linked data publishing and querying application.
The PingER ontology is an adaptation of the MOMENT (Monitoring and
Measurement in the Next Generation Technologies) ontology [SALVADOR. et al.,
2010; RAO, 2010], a core ontology which conceptualizes the networking performance
measurements domain.
The MOMENT ontology is complex and generic in the way it contemplates the
main characteristics referring to network measurement. This generality of the ontology
enables it to be adapted to many different network measurement scenarios, including the
PingER domain. However, since the ontology is so generic, the ontology fails in
representing PingER reality. Additionally, the ontology does not aim to minimize the
number of triples generated, which make it harder to process a large amount of data.
Thus, it was decided not to reuse the ontology exactly as it is, but, instead, to reuse its
concepts and ideas as basis to build an ontology more specialized for the PingER
domain, which could better support data analytical processing. The current version of
Pinger ontology has been implemented and used to publish PingER data, that can be
accessed from a SPARQL endpoint2.
Figure 1 shows an overview of the generated model. In the center of the ontology
is the main superclass, which is the Measurement class, representing the process of
acquiring measures. Measurement relates to the following classes, in order to qualify the
measurement: Metric, through measuresMetric relation to specify which network metric
is being measured; MeasurementParameters which can be specialized in PacketSize,
through hasMeasurementParameters relation, to specify the measurement attributes;
DateTime through hasDateTime relation, to specify the time interval in which the
1
https://www6.slac.stanford.edu/
2
http://pingerlod.slac.stanford.edu/sparql
measurement was made; SourceDestinationNodes, which represents the Internet Links
and is related to two types of Network Nodes, the one which performs the role of
monitor node, sending the ping signal, and the other which performs the role of
monitored node, receiving the ping. The relation is made through PingER-
ont:hasSourceNode and PingER-ont:hasDestinationNode relations, respectively.
Figure 1. PingER Ontology (SOUZA et al., 2014)
To define the parameters of time (when the measure was taken) and space
(where the network nodes - NetworkNode - are located), concepts extracted from Time
[HOBBS & PAN, 2006] and Geonames [VATANT & WICK, 2012] ontologies were
used.
3.2. M-OPL Application to PingER Network Performance Measurement
In order to apply M-OPL to the network performance measurement domain, we used the
patterns depicted in gray in Figure 2, applied in the order indicated by the darker lines.
The process was defined in the paper that originally presented M-OPL [BARCELLOS et
al., 2014]. Figure 3 shows a fragment of the resulting ontology.
Figure 2. Application order of modeling patterns
Figure 3. Derived Ontology
The first pattern applied was MEnt (Measurable Entity), which has been
extended to consider the type of measurable entity relevant to the domain, an Internet
link. This is the current Measurable Entity Type being monitored by the PingER project,
but others could be considered and then included.
After using the pattern MEnt, two patterns were applied: TMElem (Types of
Measurable Elements) and Mea (Measures). In pattern TMElem, we could identify the
Measurable Elements considered by the PingER ontology structure and characterize
them as Directly Measurable Elements (elements that do not depend on others to be
measured) or Indirectly Measurable Elements (elements that depend on other sub-
elements to be measured).
Examples of Directly Measurable Elements are Duplicate Packets and Packet
Loss, as they result from counting the associated events. Indirectly Measurable Elements
depend on sub-elements to be measured, and in PingER case they include Directivity,
Minimum Round Trip Delay, Conditional Loss Probability, Mean Opinion Score and
Average Round Trip Time, among others. Round Trip Time or RTT, for example, is
related to the distance between the nodes plus the delay at each hop along the path
between them.
The Mea pattern was used as it was defined in M-OPL. In this pattern, a
Measure quantifies a Measurable Element, characterizing a Measurable Entity Type.
Hence, the measure number of packets quantifies the measurable element packet loss
that characterizes the measurable entity of type Internet Link. But to better define an
Internet Link it was necessary to extend the M-OPL, adding the concept of
NetworkNode, which was related to Internet Link through hasSourceNode and
hasDestinationNode relations, employed according to the role that the node is
performing during measurement.
As the pattern TMElem was used, the pattern TMea (Type of Measures) also
had to be used to characterize a Measure as a Base or Derived Measure, which serves to
quantify Directly and Indirectly Measurable Elements, respectively. This pattern was
also applied exactly as it was defined in M-OPL.
After using the Mea pattern, three paths were followed in parallel. The first led
to the Measurement Units & Scales group, the second to the Measurement Procedures
group and the third to the Measurement Planning group. In the Measurement Units &
Scales group, as it is important for the domain in order to model the units and scales of
measures, the pattern MUnit&Scale was used, as it was defined in M-OPL.
In the Measurement Procedures Group, as it is not important for the domain to
detail the data collection procedures according to the different types of measures, only
was used, in this group, the pattern MProc, as it was defined in M-OPL.
In the Measurement Planning Group, the first pattern used was INeed. For
example, in our case, Know the variability of service could be considered as an instance
of Information Need. It could be used to indicate the achievement of the Measurement
Goal, which was defined in PingER domain as Check the network quality. Although not
represented in the fragment of Figure 4, measurement goals may be composed or
simple. In this case, Check the network transfer capacity could be a sub-goal of the
composed measurement goal Check the network quality.
The next pattern used after INeed was MPI-MP (Measurement Planning Item –
Measurement Procedure), which was applied in the same way that it was defined in the
M-OPL. Finally, after addressing a Measurement Planning Item, the Meas
(Measurement) pattern of Measurement & Analysis group was applied. In the Meas
pattern, Measurement is executed based on a Measurement Planning Item and adopting
a certain Measurement Procedure. It measures a Measurable Element of a Measurable
Entity applying a Measure. The result is a Measured Value, which refers to a value of a
measure scale.
To be able to represent temporal aspects in Meas pattern, TimeOfMeasurement
was added as property of the relator Measurement (actually a property of the Event
giving rise to the Relator). In UFO, Relators are derived from Events, which are
temporal based constructs.
Making a brief comparison between the ontology derived from the application of
M-OPL and the original PingER ontology, it is possible to note that the original version
of PingER ontology is more focused on treating the particular domain concepts,
representing only partially the semantics of measurements. M-OPL includes a general
knowledge about measurements, applicable to different situations. By applying the M-
OPL, these generic classes can be specialized according to the situation being
considered. For example, in our scenario, the main focus of network measurements was
performance evaluation, considered as quality measures associated to the network (using
the Ping procedure as in the SLAC laboratory). However, by specializing M-OPL
classes, we can use further grouping of goals and measures, and represent network
evaluations other than related to performance/quality, like network reliability measures
(which would include Medium Time Between Failures – MTBF, Gracefull Degradation,
Recovery Time after Failures, Medium Time to Repair – MTTR, among others)
[BALTRUNAS et al, 2014].
Considering the main patterns proposed in M-OPL and comparing with the
classes in original structure of PingER ontology, it is possible to note that some of these
patterns are already somehow represented in the original ontology structure. The Metric
class is similar to the Types of Measurable Elements pattern (TMElem), since it is
possible to represent the Measurable Elements through this class. But is not possible to
distinguish the Measurable Elements which depend or not on others to be measured
through this class, then it was necessary to add these new concepts in the ontology, by
adding the Directly Measurable and Indirectly Measurable Elements.
The Unit class is similar to the Measurement Units and Scales pattern
(MUnit&Scale), since it is possible to represent the units in which measures are
expressed through this class. But it is not possible to represent in the original ontology
the scales for measures which are partitioned according to the units, so it was necessary
to create a new class to represent the Scale element and relate it with the Measure Unit
class.
The Measurement class is similar to the Measurement pattern (Meas), since it
functions, like in M-OPL, as a Relator, connecting the classes involved in the
measurement process. However, the measurement process of the original ontology does
not consider a Measurement Planning Item neither a Measurement Procedure, so it was
necessary to create new classes to represent these elements and relate them with the
Measurement class.
By using a generic conceptualization, the derived ontology allows interoperation
with other complementary domains, which is particularly interesting for Semantic Web
applications and publication of LOD.
4. Related Work
In the literature, two works were found applying the M-OPL. In the original proposal of
M-OPL [BARCELLOS et al., 2014], it was used to build a Software Measurement
Ontology (SMO), with a very straightforward application of the patterns. In
[FRAUCHES, 2014], knowledge about the measuring process and the vocabulary
adopted in the process described in M-OPL were used, as a basis for defining an
approach for obtaining indicators from open data. The approach proposes a set of
activities that must be performed from established measurement goals, to organize the
data from an open database and to extract indicators that provide useful information for
decision making.
Applications of OPL in different domains have also been presented, most of
them confirming the possibility of reuse of the proposed patterns, and the usefulness of
the accompanying guiding process for their application. There is, though, still a lack of
cases where the derived domain ontologies have been implemented and where the
patterns have been directly imported and adapted using an existing modeling tool. The
reuse of model fragments is already supported by the OLED3 (OntoUML Light Editor),
but currently focusing on general patterns and anti-patterns included in its underlying
library.
During the application of Meas in the PingER domain it was not evident how to
explicitly represent the dimensions that qualify a measure, such as the time dimension
and geographic location dimension, which could facilitate the visualization of possible
analytical perspectives. But, in fact, considering M-OPL and its domain ontology
derivations, it does not seem reasonable to contemplate a multidimensional structure,
similar to the representation of the Data Cube Vocabulary [CYGANIAK; REYNOLDS;
TENNISON, 2014], where facts (measurements) and dimensions (associated concepts)
are at the core of the model. Although recognizing the long-term importance of
multidimensional models for analytical processing (and, of course, for exploration and
aggregation of measurements or statistical data) they serve a different purpose: to make
explicit the analytical possibilities associated to the data. But, as such, this type of
representation do not constitute a real conceptual model associated to a domain, as it
does not usually support the representation of existing relations among concepts, their
interdependences and other rules that constitute the rich semantics of the real world
conceptualizations.
From the solution found to represent the time and location aspects associated
with Measurement in PingER domain, it is possible to conclude that M-OPL can
3
https://code.google.com/p/ontouml-lightweight-editor/
represent the temporal aspect, treating it as property of the Measurement Relator. But
for other characterizations related to Measurement, it seems a better solution to
represent them as new concepts, extending some of the patterns proposed in M-OPL.
5. Conclusion
In this paper, we have presented the application of M-OPL to the network performance
measurement scenario, in order to derive a new version of PingER ontology [SOUZA et
al., 2014] so that we could take advantage of the semantic richness of measurements and
their associated concepts when interoperating PingER data with other data as linked
open data on the Web.
The benefits of applying M-OPL brought to the development of the new version
of PingER ontology were: (i) decreasing the possibility of inconsistencies and
ambiguities, since the basic patterns of the M-OPL have been developed following a
largely explored theory based on UFO; (ii) acceleration of the ontology development
process, as the patterns application process has proven to be effective and easy to use;
(iii) as already stated previously, alignment to the core modeling patterns proposed for
the measurement area can facilitate the future publication and linkage of the PingER
data with related data sources in the Web.
As future work, we are already experimenting with this new version of the
ontology, and we expect to evidentiate that the addition of semantic expressiveness
brought to the model can lead to more sophisticated and intelligent applications,
compensating the inherent increase of complexity of the ontology structure. Also,
further discussion on the multidimensional characteristics of measures would be
interesting, and what would be the best representation or derivation from a rich
conceptualization such as the one already contemplated by M-OPL constructs.
References
Baltrunas, D., Elmokashfi, A., Kvalbein, A. (2014) “Measuring the Reliability of
Mobile Broadband Networks”. In: Proceedings of the 2014 Conference on Internet
Measurement Conference, p. 45-58, New York, NY, USA.
Barcellos, M. P., Falbo, R. A., Frauches, V. G. V. (2014) “Towards a Measurement
Ontology Pattern Language”. In: 1st Joint Workshop ONTO.COM/ODISE on
Ontologies in Conceptual Modeling and Informations Systems Engineering co-
located with 8th International Conference on Formal Ontology in Information
Systems, Rio de Janeiro, Brazil.
Cottrell, L. (2001) “Internet End-to-end Performance Monitoring”, Available in:
http://www-iepm.slac.stanford.edu. Last Access: July 2015.
Cyganiak, R., Reynolds, D., Tennison, J. (2014) “The RDF Data Cube Vocabulary”. In:
W3C Recommendation. Available in: http://www.w3.org/TR/vocab-data-cube/. Last
Access: July 2015.
Falbo, R. A., Barcellos, M.P., Nardi, J. C., Guizzardi, G. (2013) “Organizing Ontology
Design Patterns as Ontology Pattern Languages”. In: 10th European Semantic Web
Conference - ESWC 2013 - Semantics and Big Data, Montpellier. The Semantic
Web: Semantics and Big Data - LNCS, 2013, v. 7882. p. 61-75, Springer.
Finkelstein, L., Leaning, M. S. (1984) “A Review of the Fundamental Concepts of
Measurement”. In: Measurement, v. 2, issue 1, p. 25-34, January-March 1984.
Frauches, V. G. V. (2014) “Uma Abordagem Baseada em Ontologias para Obtenção de
Indicadores a partir de Dados Abertos”, MSc. Dissertation, Department of Computer
Science, Federal University of Espírito Santo, Vitória, ES, Brazil.
Guizzardi, G. (2005) “Ontological Foundations for Structural Conceptual Models”, PhD
Thesis, Centre for Telematics and Information Technology (CTIT), no 05-74, The
Netherlands.
Hobbs, J. R., Pan, F. (2006) “Time Ontology in OWL”. In: W3C Working Draft, v. 27,
p. 133, September 2006.
Rao, S. (2010) “Monitoring and Measurement in the Next Generation Technologies”.
Available in: ftp://ftp.cordis.europa.eu/pub/fp7/ict/docs/future-networks/projects-
moment-factsheet_en.pdf..
Salvador, A., de Vergara, J. L., Tropea, G., Blefari-Melazzi, N., Ferreiro, Á., Katsu, Á.
(2010) “A Semantically Distributed Approach to Map IP Traffic Measurements to a
Standardized Ontology”, In: International Journal of Computer Networks &
Communications (IJCNC), v. 2, no. 1, p. 13-31.
Scherp, A., Franz, T., Saathoff, C., Staab, S. (2012) “A Core Ontology on Events for
Representing Occurrences in the Real World”, In: Multimedia Tools and
Applications, v. 58, issue 2, p. 293-331.
Souza, R. F., Cottrell, L., White, B., Campos, M. L., Mattoso, M. (2014) “Linked Open
Data Publication Strategies: Application in Networking Performance Measurement
Data”, In: The Second ASE International Conference on Big Data Science and
Computing, Stanford. ASE BIGDATA / SOCIALCOM / CYBERSECURITY
Conference.
Vatant, B., Wick, M. (2012) “Geonames ontology”, Available in:
http://www.geonames.org/ontology/documentation.html, July 2015.