=Paper=
{{Paper
|id=Vol-1920/paper5
|storemode=property
|title=Specification and Analysis of Resource Utilization Policies for Human-Intensive Systems
|pdfUrl=https://ceur-ws.org/Vol-1920/paper5.pdf
|volume=Vol-1920
|authors=Seung Yeob Shin
|dblpUrl=https://dblp.org/rec/conf/bpm/Shin17
}}
==Specification and Analysis of Resource Utilization Policies for Human-Intensive Systems==
https://ceur-ws.org/Vol-1920/paper5.pdf
Specification and Analysis of Resource
Utilization Policies for Human-Intensive Systems
(Extended Abstract)
Seung Yeob Shin
University of Massachusetts, Amherst MA 01003, USA
shin@svv.lu
https://wwwen.uni.lu/snt/people/seung yeob shin
Abstract. Societal processes, such as those used in healthcare, typi-
cally depend on the effective utilization of resources, both human and
non-human. Sound policies for the management of these resources are
crucial in assuring that these processes achieve their goals. But complex
utilization policies may govern the use of such resources, increasing the
difficulty of accurately incorporating resource considerations into com-
plex processes. This dissertation presents an approach to the specifica-
tion, allocation, and analysis of the management of such resources.
Keywords: resource utilization policy, discrete-event simulation, model check-
ing, human-intensive systems, process modeling, process analysis
1 Introduction
Complex real-world processes with human, software, and hardware resources
integrated to perform key functions, play an important role in our society. Be-
cause access to these resources is usually limited both by their quantity and by
restrictions on their availability, contention for them is often a serious problem.
Understanding how such contention arises requires that these process specifica-
tions be sufficiently powerful and precise to specify just how process activities
make use of these resources. This is further complicated by the need to also
precisely define potentially complex and numerous resource allocation policies
that are driven by process goals, regulations, or the need to satisfy the inter-
ests of different stakeholders. For instance, hospital emergency department (ED)
processes use diverse resources, such as medical staff, beds, and devices that are
often severely limited in quantity and are restricted by policies that often con-
flict. For example, in many EDs, a patient should be cared for by the same
doctor and nurse for the entire stay, but under unusual circumstances this pol-
icy may be violated to improve efficiency in patient care. Suboptimal policies for
resolving hospital resource management conflicts can result in such problems as
overcrowding, inefficient staff utilization, and long length of patient stay.
Process resource management has been widely studied through modeling
and analysis, but most prior work has used relatively simple resource models
2
that have difficulty specifying the necessary complex resource concerns flexibly
and rigorously. This dissertation provides a framework for the specification and
analysis of resources and their utilization policies in such complex processes.
The research makes the following main contributions: (1) a precise specifica-
tion language for resources and resource utilization policies, (2) a process- and
resource-aware discrete-event simulation (DES) system that supports the per-
formance of process simulations that adhere to the specifications, (3) process-
and resource-aware model checking tools that can verify process properties and
the adherence of process simulations to these policies, (4) elaborated patient
care process models based on real-world data and domain expert knowledge,
and (5) a case study applying these approaches to validating and verifying the
soundness of resource allocations to patient care processes in an ED.
2 Resource and Resource Utilization Policy Modeling
Our resource modeling approach relies on a precise, well-defined process model.
For the ED domain, this means a detailed model of the process by which patients
are treated. We use the Little-JIL language [6] to specify this model. Little-JIL
process definitions use hierarchical decomposition, exception handling, concur-
rency and human choice to structure process activities, where each activity in-
corporates the specification of the resources needed to perform the activity.
Our resource specification is orthogonal to, and separate from, activity and
dataflow specifications. A resource is modeled as the composition of a set of
attributes and a set of capabilities. A resource’s attributes, e.g., age, experience,
model number, memory footprint, and skill level, are used in deciding which
resource instance is assigned to a requesting activity. One particularly important
attribute in an ED domain is the work shift, specifying the times when a resource
can be allocated to an activity. A resource’s capabilities are the activities that
the resource can perform. For example, a set of capabilities of a doctor (MD)
includes prescribing medications and ordering tests.
Given the existence of such resource specifications, this approach enables
specification of three different kinds of resource utilization policies: permission
constraint policies, scheduling policies, and conflict resolution policies. A per-
mission constraint policy specifies the permissibility of a resource to handle a
request as restricted by a specified guard (defined as a Boolean expression). For
example, the SameMD policy can constrain the choice of MD who can handle a re-
quest a specific patient, and the ShiftMD policy can constrain the times when an
MD can treat a patient. Schedule policies support the specification of contention
and selection policies. Contention policies specify precedence among requests.
In a hospital ED, when multiple patient care activities for different patients
require the service of more MDs than are currently available for allocation, an ap-
propriate scheduling policy is necessary to resolve the contention problem among
the requests. The selection policy complements the contention policy, support-
ing specification of precedence among the resources that are able to handle a
resource request. For instance, when a new patient arrives in an ED, there are
3
usually more than two MDs who can assess the patient. An appropriate work-
load policy can balance the workloads of the MDs. Conflict resolution policies
specify how to deal with policies that come into conflict with each other and with
the situation where two or more policies cannot be enforced simultaneously. For
instance, patient handoff in a hospital ED is specified by using a conflict resolu-
tion policy because MDs are not able to satisfy both the SameMD and ShiftMD
policies at the ends of their shifts [2, 4].
3 Dynamic Analysis: Discrete-Event Simulation
The process- and resource-aware DES framework supports the execution of sim-
ulations of processes whose activities are modeled in Little-JIL, and whose re-
sources are modeled in the resource modeling approach presented in this disser-
tation by extending JSim [3]. The extension enables the simulator to support
scenarios such as (1) allowing MDs to have varying shift constraints, (2) enforc-
ing the SameMD policy unless it contradicts the ShiftMD policy, at which point
the patient is handed off to a new MD, and (3) enabling a variety of scheduling
policies, such as handing the sickest patient first, as opposed to using the least
utilized resource first [4, 5].
The dissertation presents a novel constraint-aware resource scheduling ap-
proach that consists of three steps. First, an algorithmic control method embed-
ded within the resource-aware DES framework computes resource requirements,
such as how many of each resource must be present at each time epoch to
meet user-specified resource utilization requirements. Second, deterministic In-
teger Linear Programing (ILP) produces a resource schedule that satisfies those
resource requirements and user-specified constraints on resource utilization. Fi-
nally, the resource-aware DES computes how the resource schedule affects sta-
tistical estimates of the system’s runtime properties.
4 Static Analisys: Model Checking
This process- and resource-aware DES framework shows promise as a vehicle for
evaluating diverse resource utilization policies in a flexible manner. However, the
dynamic approach verifies only that a single simulation adheres to the resource
utilization constraints, but it is inherently unable to assure that all possible
simulations adhere to the constraints. The dissertation develops a novel approach
to analyzing the application of resource utilization policies statically.
Model checking is a static analysis technology that has long been used to
either demonstrate that all possible executions of a program adhere to specified
properties, or to identify counterexample executions that violate a property, e.g.,
FLAVERS [1]. Typically, the properties are event sequence specifications char-
acterizing program functionality. This dissertation uses this technology to stat-
ically verify that all simulation executions adhere to resource utilization prop-
erties, increasing the credibility of the simulation studies and their applicability
in practice.
4
As with other model checking tools, FLAVERS requires three kinds of in-
puts: a detailed and rigorously-defined flow graph, a property finite state ma-
chine (FSM), and a set of constraint FSMs. Given the precise specifications of
resources and their utilization policies, the process- and resource-aware model
checking approach presented in the dissertation augments the three activity-
focused inputs to FLAVERS with resource utilization specifications, rigorously
defining feasible and infeasible resource allocations and utilization policies. This
enables FLAVERS to verify that resource utilization policies must always be ad-
hered to by all possible simulation runs. As an example, the dissertation presents
a verification that a specified ED patient care process satisfies the SameMD and
shift policies. This static analysis approach has also been used to support reach-
ability analyses that have detected the presence in some example processes of
resource-driven deadlocks and conflicts among policies.
5 Results
RN Utilization (BMC) RN Utilization (New)
The dissertation presents an evalu-
100
ation of the resource modeling and 90
80
analysis approach by applying it to 70
Utilization (%)
60
50
the challenging study of a detailed 40
30
hospital ED process. The ED pro- 20
10
cess model is based on real-world data 0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
and the knowledge of a domain ex- Time (hours)
pert with extensive experience as an Fig. 1. Utilization comparison over 24
emergency department physician and hours a day: RN Utilization (BMC) RN
ED manager at the Baystate Medical utilization of Baystate Medical Center, RN
Center (BMC), in Springfield, MA, Utilization (New) RN utilization derived by
the constraint-aware scheduling approach
USA. MD and nurse (RN) resources
are scheduled simultaneously while adhering to the complex characteristics of
the ED such as time-varying patient arrivals, constraints and policies of medi-
cal providers, other hospital resource (e.g., beds) utilization, and complex pa-
tient care processes. The evaluation demonstrates that this scheduling approach
creates better staffing than existing real-world staffing in terms of balancing
resource utilization over a 24-hour period (see Fig. 1).
The DES framework provides detailed trace information about resource uti-
lization to support the dynamic analysis of different resource utilization policies.
However, this dynamic analysis is valid only when all the simulations of re-
source utilization adhere to the defined resource utilization policies. Therefore,
the dissertation also includes a demonstration that the formality of the resource
specification supports verification of resource properties such as the absence of
violations of resource utilization policies and the absence of deadlock. Given the
entire ED specifications, the static analysis approach verifies the adherence to
the SameMD and ShiftMD policies; and finds conflict among hospital policies or
detects a resource deadlock under various specification of the numbers of MDs,
RNs, beds, and patients.
5
6 Contributions
Complex societal processes integrate multiple types of resources, such as hu-
mans, software, and hardware, to support the performance of process activities.
Participation by these resources is often restricted by diverse resource utilization
policies. Further, policies may at times conflict with each other, requiring con-
flict resolution strategies that add extra complexity. Resource utilization policies
usually significantly impact system behaviors and results. Therefore, resource
utilization policies should be thoroughly evaluated and rigorously analyzed.
This dissertation presents a novel approach to create specifications of policies
that are sufficiently precise and detailed to support static and dynamic analy-
ses of how these policies affect the properties of processes that are governed by
these policies. The presented process- and resource-aware DES framework sup-
ports simulations of such processes that adhere to resource specifications. The
framework’s evaluation of hospital ED processes demonstrated that it supports
considerable flexibility in resource and utilization policy specification and pro-
vides powerful dynamic analyses. However, the approach is inherently limited
because of the impossibility of exhaustive simulations of all scenarios. To com-
plement the dynamic simulation approach, this work develops a process- and
resource-aware static analysis approach that globally verifies system properties
and adherence to resource utilization policies. In doing this, the dissertation
demonstrates how the specification and analysis framework can be effective in
guiding domain experts towards sound decisions about policies for the manage-
ment of hospital resources, while also providing rigorously-based assurances that
the guidance is reliable and well-founded.
References
1. M. B. Dwyer, L. A. Clarke, J. M. Cobleigh, and G. Naumovich. Flow analysis for
verifying properties of concurrent software systems. ACM Transactions on Software
Engineering and Methodology, 13(4):359–430, Oct. 2004.
2. P. L. Henneman, S. Shin, Y. Brun, H. Balasubramanian, F. Blank, and L. J. Oster-
weil. Using computer simulation to study nurse-to-patient ratios in an emergency
department. The Journal of Nursing Administration, 45(11):551–556, 2015.
3. M. S. Raunak, L. J. Osterweil, A. Wise, L. A. Clarke, and P. L. Henneman. Simu-
lating patient flow through an emergency department using process-driven discrete
event simulation. In Software Engineering in Health Care, 2009.
4. S. Shin, Y. Brun, H. Balasubramanian, P. L. Henneman, and L. J. Osterweil.
Discrete-event simulation and integer linear programming for constraint-aware re-
source scheduling. IEEE Transactions on Systems, Man, and Cybernetics: Systems,
47, 2017.
5. S. Shin, Y. Brun, L. J. Osterweil, H. Balasubramanian, and P. L. Henneman. Re-
source specification for prototyping human-intensive systems. In International Con-
ference on Fundamental Approaches to Software Engineering, pages 332–346, 2015.
6. A. Wise. Little-JIL 1.5 language report. Technical Report 2006–051, Department
of Computer Science, University of Massachusetts, Amherst, 2006.