Supporting Cyber Threat Analysis with Service-Oriented Enterprise
Modeling
Kees Leune
a
and Sung Kim
b
Adelphi University, 1 South Avenue, Garden City NY, U.S.A.
Keywords:
Conceptual Modeling, Threat Modeling, Service-Oriented Architecture, Service-Oriented Computing,
Conceptbase, Threat Analysis, Indicators of Compromise, IOC.
Abstract:
Today’s enterprise environment is rapidly changing with organizations adopting cloud services at record rates.
This deperimeterization of enterprise computing architectures depends on software as a service (SaaS) and
makes traditional perimeter-based defense controls less effective. We propose a service-oriented threat model-
ing approach that focuses on the perspective of a service consumer. We supplement our approach by providing
an implementation view that includes technical details of service implementations that can be queried to iden-
tify potential vulnerabilities in the system. Our approach differs from existing threat modeling methods in that
we seek to capture interactions between services in a technologically agnostic manner. This extends the appli-
cability of our model into the realm of security operations. A case study and proof-of-concept are presented
to validate our approach and demonstrate how such a model can be used to provide meaningful support for
operations engineers.
1 INTRODUCTION
Threat modeling is commonly used to denote a pro-
cess that systematically enumerates attack vectors and
potential weaknesses at design time. As such, it is a
tool widely used by cybersecurity builders. However,
the term threat modeling, as used in cyber defense, is
somewhat of a misnomer since it is generally not used
to model actual threats. Rather, cyber threat model-
ing methods are used to describe systems, and to sup-
port analysis of those systems, to identify and analyze
threats.
Many different approaches to threat modeling
have emerged in literature in the recent past. Most
of those position threat modeling as an activity that
should be performed as part of requirements engi-
neering and systems design efforts. For example,
(Dhillon, 2011) sees threat modeling as a conceptual
exercise to analyze a system’s architecture or design
to find security flaws and to reduce architectural risk.
He acknowledges that threat modeling can occur at
any time during a system’s life cycle, but recommends
it to be part of the architecture or design phase. More
recently, (Sion et al., 2019) defined threat modeling
as entailing the systematic enumeration of misuse and
a
https://orcid.org/0000-0003-1171-6410
b
https://orcid.org/0000-0002-1344-5653
attack vectors and considering their applicability in
the system under design.
As early as 2008, (Malik et al., 2008) acknowl-
edged that threat modeling should be an ongoing pro-
cess and proposed a 7-step threat modeling approach
that loops continuously. However, the approach is de-
fined at a high-level, and does not propose any spe-
cific methods by which each step should be accom-
plished. In this paper, we agree with with Dhillon and
Malik that threat modeling should be considered as an
ongoing process that continues beyond requirements
engineering and design into the deployment and oper-
ation of a full enterprise computing landscape.
Today’s environment is rapidly changing. Organi-
zations are adopting cloud services at a rate not seen
before, as 92% of organizations today have IT envi-
ronments that are at least partly in the cloud (IDG,
2020). Increased use of cloud services results in de-
creased visibility into the operations of those systems.
Consequently, traditional (perimeter-based) cyberse-
curity practices become less effective.
Employees are also increasingly removed from
the enterprise network as they complete their daily
duties from home. This trend started well before the
outbreak of the global COVID-19 pandemic, but the
move to telecommuting was significantly amplified
and accelerated by lock-down mandates and social
Leune, K. and Kim, S.
Supporting Cyber Threat Analysis with Service-Oriented Enterprise Modeling.
DOI: 10.5220/0010502503850394
In Proceedings of the 18th International Conference on Security and Cryptography (SECRYPT 2021), pages 385-394
ISBN: 978-989-758-524-1
Copyright
c
2021 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
385
Figure 1: Deperimeterization.
distancing concerns.
The third observable trend is the increased adop-
tion of transport-level encryption (TLS) to crypto-
graphically protect network communications. As
of the time of writing this article, adoption of
the HTTPS-protocol to access websites exceeds
95% (Google, 2020).
The problem caused by these three coalescing
trends is illustrated in Figure 1, which shows that
most enterprise computing security controls, like fire-
walls, intrusion detection/prevention systems, and/or
data loss prevent systems, are commonly placed at
the perimeter of the enterprise network. However,
with neither the service provider, nor the service con-
sumer present on that network, the effectiveness of
perimeter-based controls has decreased since data no
longer flows through them.
The ongoing process of deperimeterization, in
which organizations continue to rapidly adopt cloud
services, and in which employees are increasingly
mobile, makes defending the enterprise environment
a much harder process. While some organizations re-
sort to the use of virtual private networks (VPNs) to
route network traffic between service providers and
their employees through the enterprise network before
sending it back out to the Internet, we see this as a so-
lution that does not scale and will not be sustainable
in the long term. In addition, requiring employees to
establish VPN connections in order to access Internet-
based resources introduces additional attack vectors
into the protected enterprise network, thus potentially
decreasing the overall security posture of the organi-
zation. The increased use of end-to-end encryption
will continue to erode the effectiveness of perimeter-
based controls.
This realization leads us to formulate our research
problem as: deperimeterization of enterprise com-
puting architectures will continue, making traditional
perimeter-based controls increasingly less effective.
Our long-term research objective is to evolve cy-
ber threat modeling to provide meaningful support to
enterprise defenders operating in a deperimeterized
computing environment for identifying and analyzing
cyber threats. We first need to determine the goal of
threat modeling, as well as the additional complexi-
ties introduced by operating in a deperimeterized en-
vironment. Next, we divide the modeling problem
into three separate sub-problems: modeling the enter-
prise computing environment, modeling threats, and
mapping threats to the environment. Each of these
steps will be validated through a manual exercise,
based on a case study. Lastly, we explore the pos-
sibility to automate each of these steps.
This paper will focus on addressing the first of the
above objectives: how can we threat model deperime-
terized environments to adequately reflect a service-
oriented approach to software utilization? We posit
that traditional threat modeling approaches, mostly
undertaken during requirements engineering and for
systems design, can be augmented to provide mean-
ingful support for operations engineers. We propose a
model with main components that benefits the service
consumer as well as the service provider.
To demonstrate our modeling process, we present
a case study and a proof of concept of our model-
ing proposal using ConceptBase implementing the O-
Telos language. We compare our modeling approach
vis-
`
a-vis traditional modeling techniques to see how
it better describes a deperimeterized environment and
how we can query our model to identify potential
threats to the described environment. We conclude
with directions for future work, including potential
approaches to automate this process.
2 RELATED WORK
2.1 Threat Modeling
Modeling is an activity in which an analyst creates
an abstract representation of the relevant aspects of
a system or process. Modeling approaches are gen-
erally specific to their purpose. For example, a data
model provides the analyst a vocabulary and a gram-
mar to express data stored in a system and/or moving
throughout a system.
Threat modeling is a structured method used to
identify possible threats to systems (Hussain et al.,
2014). Threat modeling is not new. In an early pa-
per, (Shostack, 2007) describes the experiences of
his team at Microsoft with threat modeling going
back as early as 1999. The modeling approach de-
scribed in the paper consists of four main activities:
diagramming using data flow diagrams, per-element
threat enumeration using the STRIDE methodol-
ogy, mitigation, and validation. Other well-known
SECRYPT 2021 - 18th International Conference on Security and Cryptography
386
threat modeling methods include PASTA (UcedaV-
elez and Morana, 2015), Attack Trees (Schneier,
1999), VAST (Shevchenko et al., 2018) and OC-
TAVE (Alberts et al., 2003). All threat modeling ap-
proaches have elements in common: they define a
narrow, well-defined vocabulary to describe concep-
tual elements, and the relationships that exist between
them. Some also propose a modeling process.
To be effective, cyber threat management pro-
cesses require that defenders have an established pro-
cess in place for mapping knowledge of known cyber
threats to the current state of their infrastructure. In
the context of product development, threat modeling
should be done early in the process. When developing
threat models to analyze enterprise networks, mod-
eling must be an ongoing process. As enterprise IT
landscapes are highly dynamic in nature, and new ser-
vices are adopted and brought to end-of-life continu-
ously, models of enterprise architectures must contin-
uously be updated to ensure that they are correct and
complete representations of the actual environment.
Approaches to threat modeling can be organized
into three categories: asset-centric, attacker-centric,
or software-centric (Shostack, 2007). In a later book,
Shostack argues for the software-centric approach as
the best option as it requires understanding the com-
plexity of the models, which results in substantial
improvement in the security of the software compo-
nents (Shostack, 2014). Focusing on assets requires
mapping those assets to software to be useful in iden-
tifying threats. Focusing on attackers often does not
create reproducible results. Focusing on the software
being built and the systems being deployed best incor-
porates the understanding of threats into the overall
operation of the system.
Threat modeling involves understanding the com-
plexity of the system and identifying all possible
threats to the system, regardless of whether or not they
can be exploited (Myagmar et al., 2005). The authors
discuss threat modeling in the context of requirements
analysis, which takes place prior to software design
and construction.
The focus on applying threat modeling techniques
to software development is prevalent throughout the
literature. For example, (Torr, 2005) states that “threat
modeling an entire product is typically too complex,
whereas doing so for individual features is too sim-
plistic.
(Steven, 2010) takes a broader approach to threat
modeling. He defines it as “the process of enumer-
ating and risk-rating malicious agents, their attacks,
and those attack’s possible impacts on a system’s as-
sets. However, Steven also argues that the adoption
of threat modeling as a common practice is lagging as
the result of many security managers considering it as
expensive and difficult.
In a cloud-based environment, the natural exten-
sion is to shift to focus on services in threat model-
ing. Focusing on services incorporates the compre-
hensive model of the system as in software-centric
approaches, while considering the assets that are in-
volved in the provided services. It can represent the
system from the perspective of the service consumer.
2.2 Service-Oriented Computing
Service-Oriented Computing (SOC) was established
as a field of research when it was described by (Pa-
pazoglou and Georgakopoulos, 2003) as the com-
puting paradigm that utilizes services as fundamen-
tal elements for developing applications with a num-
ber of focus areas for further exploration. In 2017,
these authors revisited their original article and paired
up with others to publish (Bouguettaya et al., 2017),
in which service-oriented computing is positioned as
having emerged as a cross-disciplinary research field
that studies the science and technology underlying the
popularity of the IT service industry.
(Papazoglou, 2003) describes the Service Ori-
ented Architecture as a logical way of designing a
software system to provide services to either end-user
applications or other services distributed in a network
through published and discoverable interfaces and on
the premise that services are loosely coupled. (Le-
une, 2007) further identifies services as an aggregate
of atomic computational operations and/or of other
services.
The service-oriented paradigm promotes describ-
ing the services provided by a technical infrastruc-
ture abstractly, and separates the description of the
service’s capabilities and interfaces from their imple-
mentation.
This approach closely resembles how modern
enterprises are deperimeterizing their IT infrastruc-
tures, and, consequently, provides a natural reference
framework. The service-oriented paradigm enables
an analyst to reason about both the capabilities pro-
vided by a service, as well as about the technical prop-
erties of the system used to provide the service.
3 SERVICE-ORIENTED
MODELING FOR CYBER
THREAT ANALYSIS
As mentioned earlier, threat modeling is predomi-
nantly embraced during the requirements analysis and
Supporting Cyber Threat Analysis with Service-Oriented Enterprise Modeling
387
design phases of a software engineering project. We
propose extending the practice by expanding threat
modeling to include separate stages for enterprise ser-
vice modeling and for cyber threat analysis.
Furthermore, threat modeling is commonly per-
formed by service providers, but its adoption by ser-
vices consumers is lagging. Our approach aims to ex-
tend the practice of threat modeling to benefit service
consumers as well.
In this paper, we adopt a service-oriented perspec-
tive on enterprise modeling, rather than a network-
centric approach. We model services, data flows, data
storage, and authentication systems conceptually (i.e.,
technology-agnostic), and then proceed to enrich the
model using technical implementation details. The
technical details provide an operational perspective,
which is mapped to the conceptual overview that pro-
vides the design view.
Our approach provides a mechanism for enterprise
service modeling in support of cyber threat analysis
conducted by service consumers. The deficiencies of
traditional modeling techniques in effectively model-
ing services from the perspective of a service con-
sumer can be attributed, at least in part, to the shift to-
wards cloud-based services. Since modelers typically
have fewer insights into the inner workings of the
technologies used to provide software-as-a-service,
modeling approaches that depend on the availability
of such information fall short.
Most of the existing threat modeling techniques
lack adequate expressiveness and semantics to enable
reasoning about threats. Therefore, the lack of ade-
quate semantics makes the development of automated
threat derivation tools and threat model validation dif-
ficult (Mirembe and Muyeba, 2008). The most widely
used threat modeling approach, STRIDE, is based on
an enhanced version of the traditional data flow dia-
gram (DFD) (Hussain et al., 2014). The four prim-
itive terms used in a DFD are the terminal (an en-
tity external to the model), the process, the data flow
representing data-in-motion, and the data store repre-
senting data-at-rest. The models used in STRIDE ex-
tend these four primitives with a fifth: a trust bound-
ary (Shostack, 2007). But in a service-oriented archi-
tecture, many of these concepts no longer apply or are
inaccessible to the analyst. The priority should be on
the capabilities of the services offered and the tech-
nical properties of how they are provided (Leune and
Kim, 2020).
Before we can establish a modeling approach, we
must identify the goals that we set out to achieve.
We believe that using threat modeling during require-
ments engineering and for systems design can be aug-
mented by evolving it to provide meaningful support
to enterprise defenders operating in a deperimeterized
computing environment.
The complexity of the discipline of secure systems
engineering requires automated tool support where
the precise definition and specification of a threat be-
comes a prerequisite (Rouland et al., 2020). Specifi-
cally, we intend to develop a method for mapping cy-
ber threat intelligence to enterprise service models in
a meaningful, and eventually, automated way. By do-
ing so, enterprise defenders will expand and improve
their situational awareness, and be able to better rec-
ognize and stop attacks.
Threat models based on services have been ex-
plored by (Kazim and Evans, 2016). They describe
(cloud) services as having four generic properties:
data storage, data processing, data transfer, and au-
thentication. Their approach uses these four proper-
ties as the basis for analyzing a select number threat
that apply to cloud services. For each threat, they
identify a threat actor and a possible method of at-
tack. While this approach can be useful, it falls short
of comprehensively analyzing the full computing en-
vironment.
Most modeling techniques incorporate knowledge
of potential threats, rather than knowledge of actual
threats. In part, that is caused by a relatively broad
approach to modeling. For example, consider a com-
posite service consisting of a web service that is ac-
cessed from a browser via HTTP/1.1 over a TLS 1.3
channel, and of an administrative service that is ac-
cessed via the secure shell protocol. Under traditional
modeling techniques, such a service would simply be
modeled as a collection of processes and data flows,
without the semantic content that would be available
otherwise (Shostack, 2007). In particular, traditional
modeling approaches use data flows merely to show
the existence of and the direction of the flow of data
between structural components (Shostack, 2014). Our
approach seeks to address this deficiency by repre-
senting how data flows between services.
We first look at defining the service as the funda-
mental concept on which this approach is based. We
then discuss how services can be combined and ex-
tended through orchestration practices. Lastly, we ex-
plore service interaction, as a basis for security anal-
ysis.
3.1 Model Semantics
Prior to defining the syntactic elements of our threat
modeling approach, we introduce its main compo-
nents and explain their semantics.
As shown in Figure 2 The central concept in our
approach is the Service. Services contain data at rest,
SECRYPT 2021 - 18th International Conference on Security and Cryptography
388
Figure 2: Model Overview.
or they receive data flows via flow channels that rep-
resent data in motion. The model is described at
two levels of abstractions: a conceptual design view,
which describes technology-agnostic properties of the
model elements, and a specific implementation view,
which provides details about their implementation.
3.1.1 Services
Definition 1 (Service). A Service provides a unit of
work.
From a service consumer’s view, a service is often
considered as a self-contained black box that logically
provides business functions. This conceptual view al-
lows the consumer to reason about the functions pro-
vided by a service, without having to also consider its
technical implementation details and service-delivery
mechanism. This view, which we refer to as the de-
sign view, is most useful when analyzing business
processes, assessing overall system designs, and de-
termining data flows into and out of the service.
However, service models used for security anal-
ysis cannot ignore implementation details. Specific
knowledge concerning application software, oper-
ating systems, networking configurations, etc., are
needed to conduct in-depth security analysis. Most
indicators of compromise that are obtained from
sources of cyber threat intelligence will provide such
details.
We acknowledge this by also defining a sup-
plemental implementation view, which describes the
technical details of service implementations. Services
can directly provide functionality, but they can also
rely on other services to provide their functions. This
is known as service aggregation or service composi-
tion.
The design service view extends the work of
Kazim and Evans referenced earlier by acknowledg-
ing that internal service properties include details
about the nature of data stored and processed by the
service, and the mechanism by which it is stored, as
well as details pertaining to data transfer and authen-
tication.
Distinguishing the design view from the imple-
mentation view leads to several items of interest.
First, the comparison of actual behavior and intended
behavior allow for the automated detection of devia-
tions between the two. Any service that is used in a
way that was not expected should be considered for
further investigation. Second, the operational behav-
ior of a system can be observed through analysis of
event logs, such as authentication logs, access control
logs, etc.
3.1.2 Data Flows and Flow Channels
Definition 2 (Data Flow). Data Flows represent data-
driven interactions between services and other ser-
vices and/or actors playing roles of Service Providers
or Service Consumers. Data flows are described in
the design view.
Services do not exist in isolation and are expected
to interact with other services and/or with individual
actors in specific roles. Data flows have long been
used to represent such data-driven interactions, and
we adopt the concept to represent interactions be-
tween services as well.
Services are provided by a service provider and
consumed by a service consumer. Both are repre-
sented in our model as roles played by specific actors.
Knowing what entities interact with a service, and in
what capacity they do so, supports investigations and
may provide the ability to attribute specific threats to
threat actors. For example, the Diamond Model for
Intrusion Analysis (Caltagirone et al., 2013) explic-
itly identifies actors in the roles of victim and/or ad-
versary. Including information regarding these actors
facilitates mappings to threat intelligence in a later
stage.
Data flows are defined at a conceptual level and, if
known, have a source and a destination.
Definition 3 (Flow Channel). A Flow Channel de-
scribes implementation-specific properties of data
flows. Flow channels are described in the implemen-
tation view.
The channel through which data flows is repre-
sented as a flow channel and a mapping between the
data flow and its flow channel is maintained. Flow
channels are described by properties, which are key-
value pairs that capture protocols used, protections
added, and any other properties that may be deemed
relevant for analysis. For example, common flow
channels properties include proto=https-over-tcp,
encryption=tls1.3, port=443, etc. Flow channel
properties can be made as specific or as fine-grained
as needed.
Supporting Cyber Threat Analysis with Service-Oriented Enterprise Modeling
389
4 VALIDATION
We validate our approach by manually modeling a
case study, and then implement the model. Since
we begin with manual validation, the case study is
intended to be of relatively low complexity. In fu-
ture work, we will automate sections of the modeling
mechanism and expand testing with additional larger
case studies.
4.1 Case Study
The case study revolves around a hypothetical orga-
nization providing consultancy services to its clients.
To do so, it relies on several cloud-hosted services.
The organization’s employees all telecommute from
various locations around the United States. Employ-
ees communicate among themselves via a messaging
app, via email, and via video chat. Client contact is
generally done via email and video chat only. All
three of these services are adopted as software-as-a-
service (SaaS) and are accessible as web applications
and as mobile apps.
The organization manages its billing and time
planning through a unified time management plat-
form, which allows a project manager to define tasks,
and for consultants to log hours to those tasks. The
system is also used to generate billing information.
There is a small corporate office, which is staffed
by the CEO and her executive assistant, as well as a
small sales staff. The staffers in the corporate offices
store their work on a self-hosted file storage device.
Remote employees can access the storage by connect-
ing to a virtual private network server, which is also
hosted on-premise.
4.2 Proof of Concept
We have implemented a proof-of-concept using Con-
ceptBase, a multi-user deductive database system im-
plementing the O-Telos language. It is a powerful
tool for metamodeling and engineering of customized
modeling languages (Jarke et al., 1995). As such, it is
an ideal tool for the task at hand. The ConceptBase
version used for this proof-of-concept implementa-
tion is designed as a client/server application.
The choice to use the Telos object model was
primarily motivated by our desire to unambiguously
define core modeling concepts, to be able to reason
about them, and to be able to query the model. We
found this ability in the O-Telos language, which pro-
vides facilities for constructing, querying, and updat-
ing structured knowledge bases (Mylopoulos et al.,
1990).
Service in Class with
attribute
description: ServiceDescription;
implementation: ServiceImplementation;
association: Role;
extends: Service
end
ServiceDescription with
attribute
dataflow: DataFlow;
data: DataElement;
authentication: Authentication
end
ServiceImplementation with
attribute
channel: FlowChannel;
storage: DataStorage;
authenticationMechanism:
AuthenticationMechanism
end
Listing 1: Model Definition Fragments.
Further benefits of using the O-Telos implemen-
tation provided by ConceptBase include the ability to
query the model using a convenient logic-based syn-
tax, a clear frame-based representation format, and
the availability of a graphical browser that lets a user
easily manipulate constructs. ConceptBase also en-
forces integrity constraints, ensuring that models are
logically consistent at all times.
An O-Telos database has two primary represen-
tation mechanisms: a logic representation based on
proposition logic, and a frame representation that
more closely resembles a typical programming lan-
guage syntax. All expressions included in this paper
are written in frame syntax to ensure greater readabil-
ity.
4.2.1 Base Model
We begin building out the base model using Concept-
Base’s O-Telos frame syntax, and define a service as
shown in Listing 1.
A Service is an instance of a generic Class
and has four typed attributes: description is
typed as a ServiceDescription, implementation as a
ServiceImplementation, etc.
The ServiceDescription contains a description
of the service’s authentication mechanism, its data
storage mechanism, and incoming and outgoing data
flows. Since O-Telos attributes can contain mul-
tiple values, any descriptive element can describe
multiple instances. The elements contained in the
ServiceDescription represent the design view.
SECRYPT 2021 - 18th International Conference on Security and Cryptography
390
Figure 3: Model.
The implementation view is captured by the
ServiceImplementation, which refers to the imple-
mentation details of the data flow channel, how data is
stored, and what authentication mechanisms are used.
Using ConceptBase’s GraphEditor, we can visual-
ize the full model as shown in Figure 3. Rectangles
represent named objects, black arrows with a solid
head represent named attributes, and magenta open-
headed arrows represent generalization relationships.
The graph defines the central Service concept
as above, but also contains definitions for the other
key modeling primitives. Specifically, it defines that
a ServiceDescription is associated with a Service
through its description attribute. The Service De-
scription contains DataElements that are stored by the
Service, and it describes the nature of the DataFlows
and Authentication requirements. DataElements are
described by properties. DataFlow can be defined as
having a source Service and a destination Service, as
well as a brief description of the flow. Note that at this
level, no technical details are included.
Likewise, the ServiceImplementation consists of
a description of the mechanism used for DataStorage
and of the mechanism by which the Service commu-
nicates with other Services. The model also maps
the service’s descriptions to their implementations.
For example, when considering the definition of the
FlowChannel, which describes the technical properties
of the channel through which a dataflow travels, the
mapping to that DataFlow is recorded as well.
4.2.2 Case Study Implementation
To support the proof-of-concept implementation of
the model, we once again use ConceptBase. How-
ever, rather than defining the model at a class level, we
focus on instance levels. In this example, we imple-
ment the MessagingService and associate it with its
MessagingDescription in ServiceDescription with
dataflow
employeeMessagingFlow: MessagingFlow
data
messagingData: MessagingData
authentication
sso: SingleSignOn
end
MessagingWebChannel in FlowChannel with
dataflow
dataflow: MessagingFlow
property
protocol: HTTPS;
tcp: TCP443
end
MessagingImplementation in
ServiceImplementation with
channel
webChannel: MessagingWebChannel
authenticationMechanism
onPremSSO: OnPremSSOServer
end
MessagingService in Service with
description
description: MessagingDescription
implementation
implementation: MessagingImplementation
association
usedByClient: Client;
usedByEmployee: Employee
end
Listing 2: Case study definitions.
ServiceDescription and its ServiceImplementation.
Due to space limitations, we only include a partial
representation of the case study in this paper. The
definition of the messaging service, as is used by both
employees and clients to interact with each other, is
captured in figure 4. A sample of corresponding O-
Telos frames in provided in Listing 2.
The ServiceDescription describes the
MessagingService as one that is used by Client and
Employee, which are instances of the ServiceConsumer
Role. The service description identifies one data
flow (MessagingFlow) and one primary data element
(MessagingData). It also describes that the service
may only be accessed after a service consumer
authenticates using SingleSignOn.
The technical details describing the service are
captured in its ServiceImplementation, which de-
scribes a MessagingWebChannel through which the
MessagingFlow will travel. The channel uses the
HTTPS protocol via its default TCP port, 443.
The implementation details also describe the ser-
vice as using an on-premise single sign-on server
(OnPremSSOServer).
Supporting Cyber Threat Analysis with Service-Oriented Enterprise Modeling
391
Figure 4: Case study.
4.2.3 Using the Model for Analysis
Capturing a computing environment in the model-
ing approach outlined in this document helps an an-
alyst visualize the interactions between services and
actors. It also provides a mechanism for answering
questions. For example, on December 12, 2020, Fire-
Eye reported a security problem that was related to its
SolarWinds software (Thomson, 2020; Zetter, 2020).
Unit42, the Threat Intelligence Unit of Palo Alto Net-
works provides several publicly available Indicators
of Compromise that are likely associated with the
breach (jadefeather, 2020). Specifically, the IOCs
contain suspicious Internet domains, as well as file
hashes to consider.
Developing an O-Telos query for this situation is
relatively easy, as domain names are associated with
flow channels.
SolarWindsQuery in QueryClass isA Service with
constraint
r: $ exists si/ServiceImplementation
(this implementation si)
and exists fc/FlowChannel
(si channel fc )
and exists fp/FlowProperty
(fc property fp )
and (
(fp in FlowChannelDestination) or
(fp in FlowChannelOrigin)
) and (
(fp domain "databasegalore.com") or
(fp domain "solartrackingsystem.net")
) $
end
The query uses the logic to limit the output
to only those services that match the constraints
specified above. First, the inheritance specifica-
tion (isA Service) ensures only Services are returned.
Next, we limit the output set to only those services
that have a service implementation that contains a
flow channel with an origin or a destination domain
that includes at-risk domains.
Using this logic, it becomes trivial to perform
threat analysis by querying the model against any
property to identify specific services.
In a more conventional threat modeling approach,
for example, one based on STRIDE-analysis of data
flow diagrams, this interaction would have been dif-
ficult to capture. Typically, external systems like So-
larwinds network management equipment, are repre-
sented as terminals, with limited information repre-
sented about them. Interactions between that termi-
nal and the system for which the threat model would
be developed would primarily focus on that system’s
expected functionality and not be readily usable for
analysis such as shown above.
5 DISCUSSION
When DFD-based threat models are used during re-
quirements analysis and to design software-based sys-
tems, interactions with third-party systems are typi-
cally modeled as terminals. Terminals are generally
not described in detail, as they are only considered as
points-of-contact with the environment in which the
newly developed system will function. As such, only
data flows to and from the terminal are modeled. In a
deperimeterized scenario, such reasoning is no longer
sufficient, and it is often necessary to capture addi-
tional details about such systems. SolarWinds, as a
network management tool, would be modeled as such
an external terminal, if it were modeled at all. Our
approach, which is able to capture additional relevant
aspects of such interactions, would be more suited for
analysis.
During the implementation of the case study, it be-
came clear that manual model creation and analysis
does not easily scale. To that effect, it will be neces-
sary to explore how modeling efforts can be supported
by analyzing electronic data sources. It is easy to
imagine a situation in which the implementation view
can be generated and/or supported with observations
derived from sources like netflows, application logs,
or authentication logs, while the design view contin-
ues to be primarily analyst-driven.
It also became clear that even simple case studies
become too complex to be fully comprehensible by
human analysts. While the modeling primitives are
intuitive, easy to understand, and can be applied with-
out much effort, resulting models may be too large for
the cognitive abilities of most human analysts.
SECRYPT 2021 - 18th International Conference on Security and Cryptography
392
However, since the models are represented in a
computer-parseable way, they lend themselves for au-
tomated analysis. Several different approaches to
such analysis can be adopted. For example, the mod-
els provide a powerful mechanism to support deci-
sions made by analysts through their query mecha-
nism. In addition, it is conceivable that supervised
machine learning algorithms can be developed to an-
alyze the models to detect significant anomalies.
One of the more pragmatic methods by which
these models can be used is by mapping existing in-
dicators or compromise to the implementation model
views. For example, IOCs describing specific IP ad-
dresses, host names, port numbers, protocols, soft-
ware versions, etc. map relatively easily to flow chan-
nel properties, as shown in section 4.2.3.
Extending the model with role-based access con-
trol analysis techniques should be fairly easy to ac-
complish as well. The model accounts for the pos-
sibility to map actors to roles, and to associate spe-
cific service provider roles and service consumer roles
with services.
6 CONCLUSIONS AND FUTURE
WORK
The confluence of increased adoption of cloud ser-
vices, the growing prevalence of end-to-end en-
crypted network communications, and the surge in
telecommuting activities has resulted in a signifi-
cant drop in the efficacy of perimeter-based controls.
This process of deperimeterization requires threat an-
alysts to rethink how they achieve and maintain sit-
uational awareness, analyze threats, and design and
build countermeasures.
This observation supports our long-term research
objective to evolve threat modeling into providing
meaningful support to defenders operating deperime-
terized enterprise computing environments.
In this paper, we looked at the first stage of this re-
search: determining how enterprise computing land-
scapes can be described. In search of an answer,
we developed a service-oriented threat modeling ap-
proach that can be used to support threat modeling.
The central concept of the approach is to adopt
a service-oriented perspective, rather than a network-
centric approach. We model services, data flows, data
storage, and authentication systems conceptually (i.e.,
technology-agnostic), and then proceed to enrich the
model using technical implementation details. The
technical details provide an operational perspective,
which is mapped to the conceptual overview that pro-
vides the design view.
Specifically, our paper makes the following contri-
butions: We adopt a service-oriented perspective, and
specifically that of a service consumer. Most exist-
ing threat modeling methods support analysis and de-
sign of software-based solutions, while we advocate
extending the use of threat modeling into the realm
of security operations. Our approach is specifically
intended to capture interactions between services, re-
gardless of their ownership or the platform through
which they are provided. Adopting a service-oriented
focus will allow us to capture security-relevant prop-
erties, other than data flows. In the proof-of-concept
implementation, we captured data-at-rest and authen-
tication mechanisms as well. In addition to modeling
data flows, we capture the properties of the channels
through which these data flows travel. Doing so will
allow for more comprehensive analysis and facilitates
mapping to threat intelligence.
Automation will play a critical part in supporting
model maintenance and in threat analysis. Incorpo-
rating automation is the subject of future research.
REFERENCES
Alberts, C. J., Dorofee, A. J., Stevens, J. F., and Woody, C.
(2003). Introduction to the OCTAVE Approach. Tech-
nical report, Carnegie Mellon University Software En-
gineering Institute.
Bouguettaya, A., Singh, M., Huhns, M., Sheng, Q. Z.,
Dong, H., Yu, Q., Neiat, A. G., Mistry, S., Benatal-
lah, B., Medjahed, B., Ouzzani, M., Casati, F., Liu,
X., Wang, H., Georgakopoulos, D., Chen, L., Nepal,
S., Malik, Z., Erradi, A., Wang, Y., Blake, B., Dustdar,
S., Leymann, F., and Papazoglou, M. (2017). A ser-
vice computing manifesto: The next 10 years. Com-
mun. ACM, 60(4):64–72.
Caltagirone, S., Pendergast, A., and Betz, C. (2013). The
Diamond Model of Intrusion Analysis. Technical re-
port, Center of Cyber Intelligence Analysis and Threat
Research.
Dhillon, D. (2011). Developer-Driven Threat Modeling.
IEEE Security & Privacy, 9(4):41–47.
Google (2020). HTTPS encryption on the web. Technical
report, Google.
Hussain, S., Kamal, A., Ahmad, S., Rasool, G., and Iqbal,
S. (2014). Threat Modelling Methodologies: A Sur-
vey. Science International, 26(4):1607–1609.
IDG (2020). 2020 IDG Cloud Computing Survey. Techni-
cal report, idg.
jadefeather (2020). Unit42 solarstorm iocs. Technical re-
port, Palo Alto Networks Unit 42.
Jarke, M., Gallersd
¨
orfer, R., Jeusfeld, M. A., Staudt, M.,
and Eherer, S. (1995). ConceptBase A deductive
object base for meta data management. Journal of In-
telligent Information Systems, 4:167–192.
Supporting Cyber Threat Analysis with Service-Oriented Enterprise Modeling
393
Kazim, M. and Evans, D. (2016). Threat Modeling for Ser-
vices in Cloud. In Proceedings 2016 IEEE Symposium
on Services-Oriented System Engineering (SOSE).
IEEE.
Leune, K. (2007). Access Control and Service-Oriented Ar-
chitectures. PhD thesis, Tilburg University, CentER.
Leune, K. and Kim, S. (2020). Service-oriented model-
ing for cyber threat analysis. In Proceedings of the
Tenth ACM Conference on Data and Application Se-
curity and Privacy, CODASPY ’20, page 150–152,
New York, NY, USA. Association for Computing Ma-
chinery.
Malik, N. A., Javed, M. Y., and Mahmud, U. (2008). Threat
modeling in pervasive computing paradigm. In 2008
New Technologies, Mobility and Security, pages 1–5.
Mirembe, D. P. and Muyeba, M. (2008). Threat Modeling
Revisted: Improved Effectiveness of Attack. In Pro-
ceedings of the Second UKSIM European Symposium
on Computer Modeling and Simulation, pages 93–98.
IEEE.
Myagmar, S., Lee, A. J., and Yurcik, W. (2005). Threat
Modeling as a Basis for Security Requirements. In
Proceedings of the Symposium on Requirements En-
gineering for Information Security (SREIS 2005).
Mylopoulos, J., Borgida, A., Jarke, M., and Koubarakis, M.
(1990). Telos: Representing Knowledge About Infor-
mation Systems. ACM Transactions on Information
Systems, 8(4):325–362.
Papazoglou, M. (2003). Service-oriented computing: con-
cepts, characteristics and directions. In Proceedings of
the Fourth International Conference on Web Informa-
tion Systems Engineering, 2003. WISE 2003. IEEE.
Papazoglou, M. and Georgakopoulos, D. (2003). Service-
Oriented Computing. Communications of the ACM,
46(10):25–28.
Rouland, Q., Hamid, B., and Jaskolka, J. (2020). Reusable
Formal Models for Threat Specification, Detection,
and Treatment. In Reuse in Emerging Software En-
gineering Practices, pages 52–68. Springer Interna-
tional Publishing.
Schneier, B. (1999). Attack Trees. Dr. Dobb’s Journal.
Shevchenko, N., Chick, T. A., O’Riordan, P., Scanlon, T. P.,
and Woody, C. (2018). Threat Modeling: A Summary
of Available Methods. Technical report, Carnegie
Mellon University Software Engineering Institute.
Shostack, A. (2007). Experiences Threat Modeling at Mi-
crosoft. Technical report, Microsoft.
Shostack, A. (2014). Threat Modeling: Designing for Se-
curity. Wiley.
Sion, L., Landuyt, D. V., Wuyts, K., and Joosen, W.
(2019). Privacy Risk Assessment for Data Subject-
Aware Threat Modeling. In IEEE Security and Pri-
vacy Workshows (SPW), pages 64–71.
Steven, J. (2010). Threat Modeling—Perhaps It’s Time.
IEEE Security & Privacy, pages 83–86.
Thomson, I. (2020). SolarWinds releases known attack
timeline, new data suggests hackers may have done
a dummy run last year. The Register.
Torr, P. (2005). Demystifying the Threat-Modeling Process.
IEEE Security & Privacy, 3(5):66–70.
UcedaVelez, T. and Morana, M. (2015). Risk Centric Threat
Modeling: Process for Attack Simulation and Threat
Analysis. Wiley.
Zetter, K. (2020). Hackers last year conducted a dry run’
of SolarWinds breach. yahoo!news.
SECRYPT 2021 - 18th International Conference on Security and Cryptography
394