Architecture Requirements for Open Inference Networks
Priscilla Naa Dedei Hammond
a
, Fritz Solms
b
and Bruce W. Watson
c
Department of Information Science, Stellenbosch University, Stellenbosch WC, South Africa
Keywords:
Open Systems, Internet of Things, Complex Event Processing, Pattern Detection, Distributed Architecture
Systems.
Abstract:
Real-time distributed Internet of Things (IoT) systems are increasingly using complex event processing to
make inferences about the environment. This mode of operation is able to reduce communication require-
ments, improve robustness and scalability, and avoid the need for big data storage and processing. With
systems making many inferences about the environment, there is no provision for general access to these in-
ferences as well as the ability to make further inferences. Most IoT systems are closed or very limited in
their openness and discoverability because they are exist for commercial purposes in which they control all
the elements of the system. To this end, we propose the concept of an Open Inference Network (OIN): a novel
open architecture for detecting and publishing complex events at various abstraction levels in an event cloud,
i.e. the set of events consumed and produced by this system. Such systems contain three types of nodes: basic
event source, inference, and activity nodes. Inference nodes detect event patterns that may encode some mean-
ing and inject corresponding higher-level events into the event cloud. Activity nodes respond to an event by
prompting an external system to perform some action; such an action may result in outputs that appear as new
events. We consider the architecture requirements for OIN by assessing the required architectural elements
against current IoT standards. These requirements mainly consist of event description and discoverability, and
security, which together enable developers to collaboratively grow and evolve OINs. This is an intermediate
study which does not include an empirical study.
1 INTRODUCTION
In order to improve scalability, robustness, privacy,
performance and environmental impact within Inter-
net of Things (IoT) based systems computing load
is increasingly moved from server-based cloud com-
puting to distributed computing done near the event
sources. In the case of Edge Computing, the process-
ing is pushed onto the devices which communicate
events to the network (Choochotkaew et al., 2017). In
the case of fog computing, the processing is done at
various levels around the event sources including the
devices which feed base events into the event cloud,
the Local Area Network (LAN) to which the devices
whose information is being processed are connected,
on mobile devices, and anywhere on the Internet (e.g.
on cloud servers) (Perera et al., 2017).
In addition to single events conveying informa-
tion, one commonly finds that meaningful informa-
a
https://orcid.org/0000-0002-7353-1844
b
https://orcid.org/0000-0003-2845-1623
c
https://orcid.org/0000-0003-0511-1837
tion can be inferred from correlations of events (Luck-
ham, 2002). Identification of event patterns produced
by event correlations is commonly done via Complex
Event Processing (CEP) (Luckham and Frasca, 1998).
These event correlations commonly involve correla-
tions across event streams and time (Choochotkaew
et al., 2017; Cugola and Margara, 2012).
Event-driven Architectures (EDAs) where deci-
sions and actions are triggered by events are used in-
creasingly to facilitate loosely-coupled asynchronous
processing within a sensory world (Dunkel, 2009;
Chandy, 2016).
The vision of ambient computing in IoT net-
works is to primarily populate environments with self-
organizing devices which use information they re-
ceive from sensors to make inferences which form the
basis for control decisions (Ali et al., 2022). Such sys-
tems would benefit from an open reference architec-
ture based on a set of public standards. Even though a
lot of research has gone into platforms and standards
supporting pervasive computing, a complete and open
reference architecture enabling different parties to in-
tegrate their self-organizing devices within an envi-
510
Hammond, P., Solms, F. and Watson, B.
Architecture Requirements for Open Inference Networks.
DOI: 10.5220/0011088800003176
In Proceedings of the 17th International Conference on Evaluation of Novel Approaches to Software Engineering (ENASE 2022), pages 510-518
ISBN: 978-989-758-568-5; ISSN: 2184-4895
Copyright
c
2022 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
ronment of ubiquitous and ambient intelligence does
not currently exist.
In this paper, we present an intermediate study
which does not include an empirical study where we
consider the architectural requirements for an OIN
which is an open EDA facilitating incremental, co-
operative construction of a system composed of three
types of sensor, inference and activity nodes which
can be integrated external systems executing business
processes.
The system of loosely-coupled nodes is envisaged
to continuously evolve and be grown by independent
entities. The construction of such a system could
be viewed as society continuously and cooperatively
evolving an open nervous system for itself. We will
call such a system an Open Inference Network (OIN).
For an OIN to be feasible, there are a number of
critical architectural requirements covering the ability
to (a) describe the data communicated with events in-
cluding structure and relationships to its environment,
(b) discover event sources which produce events one
requires for to trigger activities or make higher-level
inferences, and (c) assess the trust level of event
sources and event source registries (d) whether an
event consumer is entitled to the information provided
by an event source.
The contributions of this paper are to:
propose the concept of a fog-based software ar-
chitecture enabling the collaborative development
an OIN,
to identify the standards required to facilitate an
OIN and
to assess which standards requirements can be met
by standards specifications currently available and
which standards need to still be specified in order
to facilitate an OIN.
2 BACKGROUND
The proposed OIN would be integrated into the IoT,
would have to be able to discover the meaning and
structure of events, and would have to be able to iden-
tify event patterns resulting from meaningful event
correlations. To this end we provide an overview
of some of the concepts around IoT, ontologies, fog
computing and CEP.
IoT is a fast growing technology providing an inte-
gration infrastructure between heterogeneous objects
(such as smart devices, smart objects, sensors, actu-
ators, Radio-frequency identification (RFID), embed-
ded computers, etc.) as well as bi-directional inter-
action between the objects and people. Objects and
environments (such as smart homes, smart cities, etc.)
have become smarter via IoT, meaning they have been
enhanced for interactions with systems and/or peo-
ple and have been computerized and equipped with
network interfaces (Rajguru et al., 2015). Physical
devices embedded with sensors and actuators inter-
connected over a network continuously gather and ex-
change data, make decisions and initiate actions in or-
der to enrich and improve the environment. These de-
cisions are based on both real-time data and historical
data (Augusto and McCullagh, 2007).
CEP is a technique used for real-time process-
ing of event streams in order to reduce event load
and to detect event patterns from which meaning-
ful higher level events are inferred. Event patterns
are specified declaratively using an Event Processing
Language (EPL) and queries are continuously exe-
cuted by an event processing or rules engine against
event streams. Event patterns may specify correla-
tions across event streams and time. This generally
requires that CEP rules engines trigger state machines
and that they make use of event caching. There exist,
a variety of EPLs as well as a number of proprietary
and open source implementations of CEP rules en-
gines.
Fog computing is defined as a highly virtual-
ized platform providing computing, storage, and net-
working services between end devices and traditional
Cloud Computing Data Centers (Bonomi et al., 2012).
It bridges between cloud servers and data devices.
The key aspects of fog computing is that intelligence
and computing power is placed on the Local Area
Network (LAN). Data is transmitted from data de-
vices to a fog gateway and then to cloud servers. Fog
computing addresses the needs of IoT environments
with the high volumes of data emitted from sensors
and devices improving the overall responsiveness and
reduce network latency. Fog computing also acts as
an agent for lightweight IoT devices by keeping the
security credentials of these devices updated.
An ontology defines objects, properties, abstrac-
tions (classes) and relationships between these. On-
tologies are formally specified to standardize domain
terminology and facilitate the use of automated rea-
soners to compute inferences.
Ontologies can be specified using either the Re-
source Description Framework (RDF) or the richer
but potentially more restrictive Web Ontology Lan-
guage (OWL). RDF is widely used to describe and
exchange information on the Web. Information is pro-
vided in the form of subject-predicate-object state-
ments called triplets (e.g. “My tea is hot”). These
are commonly stored either relational databases or
triple-stores which are optimized for the task of stor-
Architecture Requirements for Open Inference Networks
511
ing triples. Each of the concepts (“my tea”, “is”,
and “hot”) would be identified through a Uniform re-
source Identifier (URI) and the database entry for a
statement would have 3 columns each containing a
URI.
RDF Schema (RDF-S) is an RDF ontology intro-
ducing the concepts of literals, data types, classes and
properties as well as specialization relationships in the
form of sub-classes and sub-properties. In addition
to RDF-S, one can use the Web Ontology Language
(OWL) to specify an ontology. OWL is a richer lan-
guage which restricts the statements one can make in
order to facilitate more rigorous and more efficient au-
tomated reasoning. It comes in flavours (OWL-Lite,
OWL-DL and OWL-Full) of increasing expressive-
ness but reduced limitations. A valid RDF document
is not necessarily valid in OWL-Lite or OWL-DL but
is valid in OWL-Full. On the other hand, OWL-DL
guarantees computability.
3 RELATED WORK
Various studies have shown that partitioning CEP
queries and distributing complex event processing
across IoT edge devices and cloud virtual machines
reduced communication volumes and improved scal-
ability, performance, robustness, flexibility and cost
(Choochotkaew et al., 2017; Ghosh and Simmhan,
2018). (Lan et al., 2019) proposed improving ef-
ficiency by decomposing CEP queries into subtasks
distributed across fog devices. This reduces com-
plexity the complexity of individual event processing
nodes.
Both (Ren et al., 2019) and (Li et al., 2018)
conducted comprehensive surveys architecture de-
sign of on emerging distributed fog/edge computing
paradigms in IoT networks. The studies did not the
event processing aspect.
(Cugola and Margara, 2012) introduced an appli-
cation domain that demands real-time processing into
data flows based on the data content and the relation-
ship between the data in a distributed manner. They
also touched on possibly creating a rule language with
both data processing and event detection capabilities
to enhance high expressiveness in catering for uncer-
tain data and rules.
Since events are the primary concept of EDA,
they need to be described by a formal structural event
model based on a precise formalism/ontology (Schaaf
et al., 2012). The key issue of a Structural Event Mod-
els is the layered hierarchy of all event types covering
the incremental enrichment of raw sensor events into
more abstract and sophisticated domain events.
(Yemson et al., 2019; Schaaf et al., 2012; Tey-
mourian and Paschke, 2009) aimed to improve in-
ferability through semantic complex event process-
ing. However, inference based on rule definitions
with rigid expressiveness limits CEP across overlap-
ping ontologies defined for different domains.
(Shaw et al., 2009) studied an architecture where
event records are analyzed as linked data from which
event characteristics and links are inferred which are
published in RDFS+OWL descriptions.
(Valle et al., 2021) identified common issues sur-
rounding the lack of understanding in the integration
of independent, heterogeneous and distributed sys-
tems but focused a high-level topological and archi-
tectural strategies for multi-system integration.
Cognitive Event Processing (CoEP) uses artificial
neural networks to learn event patterns (Yang et al.,
2015). The authors incorporated a natural language
event constructor to increase expressiveness in event
pattern definition.
4 OIN
OIN is an open network of event publishing nodes
conveying either very fine grained information or
higher-level inferred information to the network for
consumption and higher level inference by other
nodes. Both, concrete (root) events as well as inferred
abstract (higher-level) events are thus available for
higher-level event detection. To this end three types
of nodes are published to an OIN:
Sensor Nodes feed new information about the en-
vironment into the OIN. Examples include sen-
sory or device nodes which provide event streams
of sensed information and nodes which provide
information about activities in external systems.
Inference Nodes use complex event process-
ing (CEP) to infer higher-level events from finer
grained events. Complex events produced by in-
ference nodes can be used in higher level pattern
matching rules of yet higher-level nodes.
Action Nodes are special types of inference nodes
which use detected events to trigger some action
in some external system. They may optionally
publish either the results of the action as events
or incorporate information received from external
systems within generated events which may be
consumed by other nodes to either make higher-
level inferences or trigger subsequent actions (as
in a sequential process).
For an OIN to be able to evolve within an open and
dynamic world, an OIN node must be able to config-
ENASE 2022 - 17th International Conference on Evaluation of Novel Approaches to Software Engineering
512
ure and publish itself within an OIN. To this end in-
ference and activity nodes deployed in OIN discover
event sources which they require for their rule based
inference. This requires the availability of semantic
descriptions of produced events as well as the discov-
erability of event sources producing events commu-
nicating information a node requires for its decision
making. To this end they must be able to describe
the events they need for their inference as well as the
events they produce. A semantic event description
must include:
the semantic meaning of the (possibly inferred)
event itself (e.g. a wildfire),
the meaning of complex (structured) components
of the event,
the meaning and value of each leaf element of the
event,
the relation of the event to other events (e.g. in
the context of CEP pattern matching - there was
an eruption), and
for complex events, the pattern matching rule on
a semantic level.
Being an open network where different, unknown par-
ties may produce and consume events requires that
nodes to assess whether they can trust a node which
either produces or consumes events, i.e. neither will a
node generally be willing to blindly trust an unknown
event source that the events it produces are correct,
nor is the information communicated with produced
events meant to be consumable by any node owned
by some unknown party. Event sources thus need to
publish:
their location on the internet,
the semantics and structure of the information
contained in the events they produce,
the technology used to encode the event (e.g.
JSON or XML),
the physical communication channels over which
the events are made available,
information which enables nodes to assess
whether they are willing to trust the event source
and
optionally restrictions on who may consume the
events it produces.
4.1 OIN Application
Different entities can publish new event sources, in-
ference and action nodes to the network producing
events which may be consumed by any interested par-
ties for their purposes. A software application, on the
other hand, is meant to address some specific task. An
OIN is expected to facilitate many applications and
nodes can be shared across these. A software appli-
cation leveraging OIN would contain elements which
are not part of the OIN itself like external services,
databases, user interfaces and so on.
Consider, for example, the view onto a subset of
a potential OIN depicted in Figure 1. Part of the de-
picted OIN is used within a wildfire management ap-
plication. Wildfires are common, particularly in the
Western Cape, South Africa, where huge areas are
regularly destroyed by fire leading to massive envi-
ronmental, health and safety as well as cost impli-
cations. Early detection of fires is critical to reduce
damage and cost.
In areas which have high wild-fire risk it is com-
mon to have a network of sensors measuring temper-
ature, air and soil humidity, carbon monoxide, wind
speed and wind-direction and light intensity (Natara-
jan and Manu, 2017).
The base layer is a sensor (diamonds) layer which
includes from left to right light-intensity, temperature,
soil-humidity, carbon monoxide, sound, rain, wind
(speed and direction), and industrial emission sensors.
The ovals represent inference nodes which use CEP
to infer higher level events from event streams rep-
resented by dashed lines with arrows indicating the
direction of event flow. For example, high-fire-risk
events are inferred from the fire-danger CEP node
(fire in a warning triangle) from light intensity, air
temperature and soil humidity events. On the other
hand the presence, intensity and area of a fire are in-
ferred by fire detection nodes from event streams orig-
inating from the same temperature sensors as well
as from carbon-monoxide and sound sensors. The
inferred fire description events together with wind,
soil humidity and rain events are used in conjunction
with an external fire-spread prediction service (which
has access to maps reporting combustible materials
such as vegetation, buildings, among others) to gen-
erate fire-spread prediction events. Various warning
systems could subscribe to this inferred event stream
to issue fire warnings to parties which are at health,
safety or damage risk. Furthermore, the fire-spread
prediction event stream together with the fire risk and
fire description event streams are used by an external
fire combating service to initiate and optimize the ef-
fectiveness of fire combating activities.
The concept of an application is a little arbitrary
within such an architecture. It typically involves
grouping of nodes and external services based on joint
purpose and ownership. For example the fire-related
Architecture Requirements for Open Inference Networks
513
Figure 1: A view onto part of an OIN showing sensor nodes, CEP inference nodes, activity nodes, external services and
grouping into applications.
services and nodes as well as some of the carbon-
monoxide sensors are here grouped into a fire man-
agement application. The application does, however,
feed off areas of the OIN which are not part of the ap-
plication itself (light-intensity, temperature, soil hu-
midity, sound, rain and wind sensors) and generates
event streams which can be consumed by other appli-
cations built around the OIN by potentially other par-
ties. For example, the inferred fire-spread prediction
events from the fire management application are used
in conjunction with event streams from CO, wind
and industrial emission sensors to generate air qual-
ity prediction events which can be consumed by an
air-quality warning service. The sound event streams
are not only used to detect the sounds of a fire, but
also by bird song inference nodes which use an ex-
ternal bird song identification service to generate bird
spotting events.
4.2 Limitations
Since OIN is an open community-driven network
where different parties independently deploy and re-
move different types of nodes, nodes cannot be hard-
wired to other nodes. Instead, nodes auto-configure
and auto-evolve their connectivity based on the avail-
ability of nodes which can provide streams of required
events.
IoT systems are naturally event-based systems.
EDAs have been shown to be well-suited for complex
and continuously evolving systems (Chandy, 2016).
Yet the software architectures of IoT systems are not
commonly modeled using only a subset of concepts
from EDAs.
There are a number of aspects which can be ac-
commodated within the proposed software architec-
ture but for which the architecture does not make ex-
plicit provision - i.e. it is left to either the infrastruc-
ture which deploys components into the open CEP
IoT architecture or to the logic of the nodes them-
selves. Examples include:
the deployment of nodes may be automated and
optimized by client software feeding nodes into
the architecture (Cugola and Margara, 2012)
Nodes may use learning algorithms to identify
meaningful event patterns and publish themselves
as event sources for such meaningful events
(Mehdiyev et al., 2015; Janjua et al., 2019)
5 APPROACH
In view of supporting the open, community-based as-
pect of an OIN, the core architectural concerns are
1. the ability to discover event sources which are in
a position to provide the information required for
certain inferences or activities and
2. the ability to assess the trustworthiness of a dis-
covered event source.
The assessment of standards support for an OIN will
hence focus on the discoverabilty and security with.
In this section, we discuss the methods used to
identify the architectural and standards requirements
around these quality attributes and to assess the re-
quirements coverage of currently available public
standards. This was done using the following ap-
proach.
1. Refining the architectural requirements for the
two quality requirements.
This was done by explicitly defining the objec-
tives and sub-objectives of an OIN architecture in
ENASE 2022 - 17th International Conference on Evaluation of Novel Approaches to Software Engineering
514
relation to its key quality attributes. With each
objective, its purpose, the motivation for its exis-
tence, its high-level scope of its functionality, and
its benefits, opportunities and limitations were
also defined.
2. Identifying aspects of these architectural require-
ments which need to be supported by public stan-
dards.
The aspects were identified by breaking the OIN
architecture into domain areas for which stan-
dards are required: Ontology specification, rules,
data encoding, trust assessment, access control,
conceptual data structure specification and event
source registry service (ESRS).
3. Identifying sources from which standards candi-
dates which may meet aspects of the the standards
requirements can be discovered.
This was done by investigating public standards
documentations published by standard organisa-
tions and conducting literature search on aca-
demic papers for the identified standard domains
as summarized in Table 1.
4. Associate with each standard requirement poten-
tial standard candidates which may meet the stan-
dard requirement either fully or partially.
This was done by identifying for each standard re-
quirement its domain and identifying those stan-
dards which target that domain.
5. Assessing the extend to which these standards
candidates meet the requirements.
By spanning and mapping the capabilities of ex-
isting tools that use these standards against the re-
quirements scope for each standard domain and
indicate what they can cater for the corresponding
requirement. This also includes identifying the re-
quirements for which standards still need to be de-
veloped.
6 STANDARDS REQUIREMENTS
FOR AN OIN ARCHITECTURE
The architecture of an OIN is based on the IoT and
will thus rely on the standards developed for the In-
ternet itself (TCP/IP, DNS, among others) as well
as standards used and being developed for the IoT.
These include a range of established and emerging
standards for light-weight and secure communication,
event/message publication and streaming (MQTT,
AMQP, among others), request-response messaging
as well as developing (COAP), deploying or manag-
ing nodes (TR-069 and OMA-DM) and distributing
tasks across node. For an OIN to function across tool
suites and across nodes developed and maintained by
different organizations, a set of additional standards
particularly to facilitate discoverability. This includes
standards for (a) describing primitive and complex
events, (b) encoding events, (c) assessing whether
an entity (event producer or event consumer) can be
trusted, and (d) publishing and discovering trusted
event producers (event sources or inference nodes)
which produce required events encoded in.
This section only considers standards require-
ments for an OIN. In section 7, we assess the stan-
dards availability, i.e. which standards requirements
can be met with available standards and standards
which still need to be developed.
6.1 Event Description
An event may contain event metadata common to all
events (e.g. address of the source node, the time
stamp of the event), data which is specific to the type
of event and (e.g. sensed temperature) as well as in-
ferred meaning generated by inference nodes (e.g. the
presence and attributes of a fire inferred from sensed
events). Since event types are to be published by very
different entities, event consumers must be able to dis-
cover event data structures including the meaning and
relationships between data items.
Note that schema specification standards like eX-
tensible Markup Language (XML) Schema are not
suited for an open world where the common seman-
tics could be encoded by different entities within dif-
ferent data structures. On the other hand, ontolo-
gies are specifically designed for an open world sce-
nario. They are used to share and communicate a
common understanding for different domains. OIN
require standards for specifying and querying and re-
lating ontologies. The latter includes related concepts
across ontologies including the ability to specify that
two concepts in two ontologies are equivalent. The
latter is required as different entities may work within
their own ontologies which may have semantic over-
laps with ontologies used by other, independent enti-
ties.
Additionally, an OIN will require the ability to
register ontologies used by event producers contribut-
ing sensed or inferred events to the OIN as well as ef-
ficient querying of a distributed set of related ontolo-
gies. The latter will most certainly require distributed
processing of queries (Sakr et al., 2018).
Architecture Requirements for Open Inference Networks
515
Table 1: The standard domains for OIN with corresponding public standard documentations.
Standard domain Resources
Ontology specification
– Resource Description Format (RDF) (Gibbins and Shadbolt, 2009)
– Web Ontology Language (OWL) (Bechhofer et al., 2004a)
Rules
– Rule Interchange Format (W3C RIF) (Kifer, 2008)
– Reaction RuleML (Paschke, 2014; Paschke, 2006)
Conceptual data structure spec
– Web Ontology Language (OWL) (Bechhofer et al., 2004a)
Data encoding
– Concise Binary Object Representation (CBOR)
(Bormann and Hoffman, 2013)
– Javascript Object Notation (JSON) (Crockford and Morningstar, 2017)
– eXtended Markup Language (XML) (Bechhofer et al., 2004b)
Trust assessment &
access control
– Internet X.509 Certificate(RFC5280) (Boeyen et al., 2008)
– ISO std for information security (ISO27001) (Watkins and Safari, 2013)
Event source registry
– Standard will need to be developed
6.2 Discoverability
Since an OIN is an open inference network with nodes
being added, removed and modified continuously, a
node must be able to query event producers (sensor,
inference or activity nodes) which are both trusted and
generate events containing the information required
to either make some higher level inference or trigger
some activity. To this end, nodes must be able to pub-
lish (a) its location on the Internet, (b) the physical
event stream(s) it makes available, (c) either an em-
bedded description or a link to a description of the
events published on these event streams including
the variable data (e.g. sensed or inferred values), and
(c) fixed data which does not change from event to
event (e.g. the physical location of a stationary node
or the units within which a measured quantity is pub-
lished), (d) the encoding used for the events, and (e)
credentials which can be used to assess whether the
producer is trusted by the consumer.
A query for suitable event producers would need
to specify a description of the information required
with the event and a set of constraints for the events
the consumer intends to receive. Event constraints
may be (a) constraints around the required data fields,
(b) constraints on fixed data for an event stream (e.g.
want to only receive temperature reading from nodes
located within a specific physical region), and (d) con-
straints on the entity vouching for the data (trustabil-
ity).
Furthermore, since an OIN is an open and poten-
tially global infrastructure within which a new form
of applications is being developed, the lookup service
needs to be a hierarchical and decentralized event pro-
ducer lookup service propagating information around
event sources in a way similar to what Domain Name
Systems (DNS) currently propagate domain name in-
formation across a network of hierarchical domain
name servers. An OIN thus requires a set of stan-
dards facilitating the interoperability of such an Event
Source Registry System (ESRS).
Furthermore, as event sources are modified, re-
moved and added nodes requiring events may have
to regularly update the event channels they observe
making further queries to the ESRS.
6.3 Security
Even though an OIN is an open network, aspects of
security are very important. In particular, an OIN
faces threats of (a) access to event source registry,
(b) access to event streams, (c) encryption of event
data, and (d) certificate-chain based vouching of info
(events).
7 ASSESSMENT OF AVAILABLE
STANDARDS
In this section we consider the standards requirements
from the previous section and identify currently avail-
able standards which meet aspects of these require-
ments as well aspects of the standards requirements
which no standards could be identified. Table 2 pro-
vides a summary of the domains for which an OIN
requires standards and, when available, standard can-
didates which address these requirements.
The RDF provides a basic set of elements to de-
scribe ontologies, whilst the OWL provides a much
more extensive set of elements which also enable
one to relate ontologies and to specify equivalence
relationships between concepts within different on-
tologies. Furthermore, certain constrained OWL ver-
sions (e.g. OWL-DL) enforce computational com-
pleteness and decidability facilitating practical rea-
soning algorithms including the ability to reason in
the context of ontologies containing contradictions ei-
ENASE 2022 - 17th International Conference on Evaluation of Novel Approaches to Software Engineering
516
Table 2: The domains for which OIN requires standards and available standards meeting these standards requirements.
Standard domain Standard candidates
Ontology specification
– Resource Description Format (RDF)
– Web Ontology Language (OWL)
Rules
– Rule Interchange Format (W3C RIF)
– Reaction RuleML
Conceptual data structure spec
– OWL, OWL-DL
– Web Ontology Language (OWL)
Data encoding
– Concise Binary Object Representation (CBOR)
– Javascript Object Notation (JSON)
– eXtended Markup Language (XML)
Mapping to physical
CBOR/RDF-JSON/XML serialization
Event messaging
MQTT, AMQP
Event streaming
Only tool-specific: Apache Kafka, IBM Event Stream, . . .
Standards facilitating ESRS
Currently no candidates available
Trust assessment
SSL/TLS with X.509 certificates
Access control
SSL/TLS with X.509 certificates
ther within an ontology or across ontologies. Reason-
ing algorithms facilitating distributed processing do
exist. Both OWL and RDF are standards maintained
by the World Wide Web Consortium (W3C).
Furthermore, both OWL and RDFS provide the fa-
cility to specify classes representing data structures.
These need to be encoded in reasonably efficient
forms for the OIN. The Concise Binary Object Rep-
resentation (CBOR) provides a standard for efficient
encoding of data structures. Other, less efficient alter-
natives are JSON and XML schema.
The standards required for a Event Source Reg-
istry System (ESRS) as described in section 6.2 are
currently not available and will need to be specified.
Finally, trust assessment can be done using SSL/TLS
with X.509 standards around maintaining, storing and
querying certificate chains.
8 CONCLUSION
We proposed a novel fog-based software architec-
ture and development paradigm for collaboratively
developing systems within an open inference net-
work using sensor, inference and activity nodes. Lev-
els of inference nodes incrementally make infer-
ences at different levels of abstraction and make this
higher-level intelligence available to be used for fur-
ther inference or to trigger activities within systems.
An OIN requires standards for describing ontologies
and data structures, reasoning, discovering suitable
event sources through a network of hierarchical event
source registries (an event source registry system,
ESRS), publishing and streaming events and assess-
ing trustability of event sources and consumers. We
have identified standards addressing most of these re-
quirements. Areas which do not have current stan-
dards support are the standards required around the
ESRS and event streaming.
The future work will focus on including the speci-
fication of standards required for event streaming and
ESRS, and the development of a reference implemen-
tation of an OIN which will represent the proof of
concept implementation for empirical studies.
ACKNOWLEDGEMENTS
This work is based on research supported in part by
the National Research Foundation of South Africa
(NRF) (Grant number: 106028)
REFERENCES
Ali, O., Ishak, M. K., Bhatti, M. K. L., Khan, I., and Kim,
K.-I. (2022). A comprehensive review of internet of
things: Technology stack, middlewares, and fog/edge
computing interface. Sensors, 22(3).
Augusto, J. C. and McCullagh, P. (2007). Ambient intelli-
gence: Concepts and applications. Computer Science
and Information Systems, 4(1):1–27.
Bechhofer, S., Harmelen, F., Hendler, J., Horrocks, I.,
Mcguinness, D., Patel-Schneider, P., and Stein, L.
(2004a). Owl web ontology language reference.
Bechhofer, S., Harmelen, F., Hendler, J., Horrocks, I.,
Mcguinness, D., Patel-Schneider, P., and Stein, L.
(2004b). Owl web ontology language reference.
Boeyen, S., Santesson, S., Polk, T., Housley, R., Farrell, S.,
and Cooper, D. (2008). Internet X.509 Public Key In-
frastructure Certificate and Certificate Revocation List
(CRL) Profile. RFC 5280.
Architecture Requirements for Open Inference Networks
517
Bonomi, F., Milito, R., Zhu, J., and Addepalli, S. (2012).
Fog computing and its role in the internet of things. In
Proceedings of the First Edition of the MCC Workshop
on Mobile Cloud Computing, MCC ’12, page 13–16,
New York, NY, USA. Association for Computing Ma-
chinery.
Bormann, C. and Hoffman, P. E. (2013). Concise Binary
Object Representation (CBOR). RFC 7049.
Chandy, K. M. (2016). Event Driven Architecture, page
1–5. Springer New York, New York, NY.
Choochotkaew, S., Yamaguchi, H., Higashino, T., Shibuya,
M., and Hasegawa, T. (2017). Edgecep: Fully-
distributed complex event processing on iot edges.
page 121–129.
Crockford, D. and Morningstar, C. (2017). Standard ecma-
404 the json data interchange syntax.
Cugola, G. and Margara, A. (2012). Low latency complex
event processing on parallel hardware. Journal of Par-
allel and Distributed Computing, 72(2):205–218.
Dunkel, J. (2009). On complex event processing for sensor
networks. In 2009 International Symposium on Au-
tonomous Decentralized Systems, pages 1–6.
Ghosh, R. and Simmhan, Y. (2018). Distributed scheduling
of event analytics across edge and cloud. ACM Trans.
Cyber-Phys. Syst., 2(4).
Gibbins, N. and Shadbolt, N. (2009). Resource description
framework (rdf).
Janjua, Z. H., Vecchio, M., Antonini, M., and Antonelli,
F. (2019). Irese: An intelligent rare-event detec-
tion system using unsupervised learning on the iot
edge. Engineering Applications of Artificial Intelli-
gence, 84:41–50.
Kifer, M. (2008). Rule interchange format: The framework.
volume 5341, pages 1–2.
Lan, L., Shi, R., Wang, B., Zhang, L., and Jiang, N. (2019).
A universal complex event processing mechanism
based on edge computing for internet of things real-
time monitoring. IEEE Access, 7:101865–101878.
Li, C., Xue, Y., Wang, J., Zhang, W., and Li, T. (2018).
Edge-oriented computing paradigms: A survey on
architecture design and system management. ACM
Comput. Surv., 51(2).
Luckham, D. (2002). The Power of Events: An Introduc-
tion to Complex Event Processing in Distributed En-
terprise Systems. Addison-Wesley.
Luckham, D. C. and Frasca, B. (1998). Complex event pro-
cessing in distributed systems. Technical report, Stan-
ford University.
Mehdiyev, N., Krumeich, J., Enke, D., Werth, D., and Loos,
P. (2015). Determination of rule patterns in com-
plex event processing using machine learning tech-
niques. Procedia Computer Science, 61(Supplement
C):395–401. Complex Adaptive Systems San Jose,
CA November 2-4, 2015.
Natarajan, T. and Manu, V. (2017). Integrated forest fire and
wildlife monitoring system using wireless sensor net-
work. Journal of Embedded Systems and Processing,
2:1–7.
Paschke, A. (2006). The reaction ruleml classification of the
event / action / state processing and reasoning space.
CoRR, abs/cs/0611047.
Paschke, A. (2014). Reaction ruleml 1.0 for rules, events
and actions in semantic complex event processing.
Perera, C., Qin, Y., Estrella, J. C., Reiff-Marganiec, S., and
Vasilakos, A. V. (2017). Fog computing for sustain-
able smart cities: A survey. ACM Computing Surveys
(CSUR), 50(3):1–43.
Rajguru, S., Kinhekar, S., and Pati, S. (2015). Analysis of
internet of things in a smart environment. Analysis,
4(4).
Ren, J., Zhang, D., He, S., Zhang, Y., and Li, T. (2019). A
survey on end-edge-cloud orchestrated network com-
puting paradigms: Transparent computing, mobile
edge computing, fog computing, and cloudlet. ACM
Comput. Surv., 52(6).
Sakr, S., Wylot, M., Mutharaju, R., and Danh Le Phuoc
(Author), I. F. (2018). Linked Data: Storing, Query-
ing, and Reasoning. Springer.
Schaaf, M., Grivas, S. G., Ackermann, D., Diekmann, A.,
Koschel, A., and Astrova, I. (2012). Semantic com-
plex event processing. Recent Researches in Applied
Information Science, pages 38–43.
Shaw, R., Troncy, R., and Hardman, L. (2009). Lode: Link-
ing open descriptions of events. In G
´
omez-P
´
erez, A.,
Yu, Y., and Ding, Y., editors, The Semantic Web, pages
153–167, Berlin, Heidelberg. Springer Berlin Heidel-
berg.
Teymourian, K. and Paschke, A. (2009). Towards semantic
event processing. In Proceedings of the Third ACM
International Conference on Distributed Event-Based
Systems, DEBS ’09, New York, NY, USA. Associa-
tion for Computing Machinery.
Valle, P. H. D., Garc
´
es, L., and Nakagawa, E. Y. (2021). Ar-
chitectural strategies for interoperability of software-
intensive systems: practitioners’ perspective. In Pro-
ceedings of the 36th Annual ACM Symposium on Ap-
plied Computing, pages 1399–1408.
Watkins, S. and Safari, a. O. M. C. (2013). An Introduction
to Information Security and ISO27001:2013, A Pocket
Guide, Second Edition. IT Governance Publishing.
Yang, J., Ma, M., Wang, P., and Liu, L. (2015). From
complex event processing to cognitive event process-
ing: Approaches, challenges, and opportunities. In
2015 IEEE 12th Intl Conf on Ubiquitous Intelligence
and Computing and 2015 IEEE 12th Intl Conf on
Autonomic and Trusted Computing and 2015 IEEE
15th Intl Conf on Scalable Computing and Commu-
nications and Its Associated Workshops (UIC-ATC-
ScalCom), pages 1432–1438. IEEE.
Yemson, R., Konur, S., and Thakker, D. (2019). A novel
semantic complex event processing framework for
streaming processing. In Proceedings of the 9th In-
ternational Conference on the Internet of Things, IoT
2019, New York, NY, USA. Association for Comput-
ing Machinery.
ENASE 2022 - 17th International Conference on Evaluation of Novel Approaches to Software Engineering
518