Trustworthiness Measurement Algorithm for TWfMS Based on Software Behaviour Entropy

The workflow management system (WfMS) is considered a multidisciplinary system-of-systems (SoS)-oriented software under the modern IT background of cloud computing, internet of things (IoT), big data, and advanced technologies for the future. When a WfMS encounters unexpected accidents, human intervention or offline adjustment is insufficient to be appropriated for its complex undertaken commissions and high availability requirements. Similar to real-time systems, a WfMS requires online adjustment to accommodate random changes occurring in its surroundings or software architecture under human control strategies, which may affect its functionalities or non-functionalities and transaction data consistence—and even its trustworthiness level. For example, addressing course timetabling problems [1] from the case-based reasoning (CBR) system viewpoint, to adapt to the limited classroom capacity for the ever-increasing number of students, the CBR system may change the venue to a new classroom with larger capacity to prevent basic teaching functionality from being interrupted. In addition, to adapt to enable more effective teaching, the CBR system may adjust the venue to a new classroom that better suits the preference of the teacher and students to optimise the potential teaching non-functionality. Furthermore, to adapt to achieve more consistent teaching transactions log data, the CBR system may recover or repair the transactions with a uniform state; otherwise, the complete student achievement would not be generated from the conflicting data. Although the CBR system works for education-oriented business process management tasks, i.e., a special type of WfMS, the random changes mentioned above could mislead the WfMS software behaviour and eventually directly or indirectly affect the WfMS service behaviour realised by its software, ultimately reducing the WfMS software trustworthiness and resulting in users abandoning the candidate WfMS adoption. In order to solve the abovementioned problem, we present a novel measurement algorithm based on previous research results. Our approach includes a self-automatic framework for random changes and relative algorithms, concentrating on measuring the similarity between software behaviour and its claims through black-box testing with incomplete software behaviour entropy. The remainder of this paper consists of three sections: Section 2 discusses the related work on measurement methods regarding the WfMS; Section 3 illustrates the novel measurement algorithm from two perspectives, namely the calculus implemented on the WfMS and the normal forms of the WfMS; and Section 4 discusses the proposed algorithm and summarises future research areas.

Software trustworthiness [2] is considered a non-functionality combination of software quality against the conformance degree oriented to subjective user evaluation. Against this background, the difference between user requirements and actual software behaviour will determine the trustworthiness perceived by the user. In recent decades, considerable efforts have been made to design intelligence algorithms to measure the difference between objective software or service behaviour and subjective user evaluation. In references [3,4], the authors define the similarity and design based on the genes of the firing sequences generated from transition adjacency relations (TARs). The corresponding distance measures between processes were taken as a metric to be applied in artificial processes and evaluations for clustering real-life processes. As a promising candidate for further research areas, the authors of reference [5] took into account the active feedback of evaluation data in workflow process modelling, which encompassed the entire lifecycle of workflow and enabled active real-time controlling using workflow audit trail data from three perspectives, namely process, resource, and object. In order to assist users in selecting a sufficient workflow with appropriate quality of service (QoS) to meet their requirements, reference [6] proposed a novel approach to scientific workflow retrieval with cost constraints and defined a distance measure for comparing the similarities among cost constrained graph (CCGs), through which workflow retrieval and ranking could be conducted based on similarity computations. To allow companies to respond to changing markets by creating product variants derived from different combinations of existing or new modular components, reference [7] sought to define and measure workflow modularity. The authors made three contributions: they designed two important performance measures regarding flow time and flexibility; they proposed an integer nonlinear programming (INLP) optimisation model for designing modular workflows that can be adopted for small processes; and they presented a heuristic model for the same commission adopted for larger processes. From a workflow standardisation perspective, in order to translate business data into bytes that are consumable by daily information systems with less automation difficulty and more reliability, as well as to make business competition comparisons more precise and invariable, reference [8] focused on health care data and workflow, reengineering the coding workflow and performance benchmarks, and proposed the establishment of useful tools for improving data quality, which can be used by everyone. In terms of the practical computation environment known as grid computing, reference [9] layered the clustering approach of the reusable workflow into a hierarchical model consisting of activity, event, condition-action, rule, and process similarity measures based on event condition action event condition action (ECA) theories. In the industry-integrated manufacturing environment, digital printing should eliminate redundant processes in order to shorten press runs and save costs. In order to select equipment and software for supporting such workflows, reference [10] introduced a workflow configuration tool that can elicit customer printing requirements and classified the requirements into a set of equivalent workflows according to their configuration using a compression-based dissimilarity measure (CDM) approach. The continuous improvement of business process management (BPM) is an ongoing issue for the WfMS, and in order to address this, a solid understanding of the success elements and waste experience during the same workflow process is essential. With the trend of lean production, reference [11] surveyed research and presented a conceptualisation for understanding the workflow and simultaneously occurring waste in the production of buildings through the three different dimensions of smoothness (high level of direct work), quality, and intensity. Moreover, they summarised the methods for this purpose, consisting of an integrated method of observation and self-reporting, as well as the last planner system (LPS) based on the method to measure workflow as handover of work between trades.

From an overview of the above related works, we can draw the conclusion that BPM measurement can be transformed from socialised human production to mechanised computing by means of electronic equipment. Therefore, there is a need for a comprehensive approach to connect the macroscopic and microscopic measurements systematically, thereby reflecting the user’s subjective experience of the trustworthiness of BPM according to the objective evidence of trustworthiness of the WfMS. However, in the dynamic and open WfMS runtime environment, the goal of comprehensively interpreting and obtaining the WfMS software behaviour is almost impossible to achieve. Generally, we can only obtain partial software behaviour, which represents incomplete trustworthiness of the entire service QoS without sufficient preciseness when the parameters regarding other software behaviour are uncertain [12]. Accordingly, we first present a quality management system (QMS) [13] of maximum entropy (QMSOME) derived from the trustworthy WfMS (TWfMS) framework, which is a thermodynamics-related interpretive model, as the virtual mirror of complex real-time business processes of the underlying information systems of organisations. Secondly, we present a novel recursive measurement algorithm for WfMS trustworthiness by means of a hybrid method derived from large-scale software black-box testing, as illustrated in the following sections, which is also an application of the principle of maximum entropy (POME) or maximum-entropy principle (MEP) [14,15].

In line with the related works discussed above, as a special type of QMS for workflow quality management, WfMSs experience multiple loops consisting of re-engineering and/or reorganisations. This process improves the WfMS into a high-level order—with relatively low entropy of information systems—from low-level disorder—with high entropy of information systems in the long term— along with consistently maintaining its trustworthiness as defined by users. This cycle, which covers the entire life of the WfMS, is also known as the resilience engineering (RE) [16] model and is illustrated in this section. Thereafter, we introduce its definition, including integrated manufacturing business processes, and the translation of these into models. Prior to proposing the RE model for the WfMS, we illustrate our thoughts regarding the TWfMS according to the standard WfMC reference model with the extension of methods and mechanisms, which is the fundamental goal of RE. Considering that the implementation of RE for the WfMS is ultimately imposed on the WfMS components, we first provide the preliminary component definition for the WfMS. Secondly, as the fundamental goal of WfMS RE is to assure its trustworthiness as perceived by users, we provide the trustworthy component definition of the WfMS. Thirdly, in order to certify WfMS recovery as its common function under RE, we present a normal form (NF) set [17] as a so-called paradigm to label the WfMS trustworthiness level at runtime.

In order to address the measurement problem of WfMS trustworthiness, based on prior research [17,18,19,20,21], this paper has proposed a novel algorithm for WfMS trustworthiness based on the TWfMS framework mechanisms in an uncertain environment with incomplete software behaviour test cases, which means that the deterministic entropy of services or its underlying software behaviour is partial. Similar to BPM, we can consider the entire WfMS lifecycle as a group of long-term processes, which we categorise into three aspects: the ‘as-is process’ in the build-time stage, the ‘to-be process’ in the runtime stage, and the ‘agile-consistent process’ in maintenance time. This study focuses on the measurement algorithm of the first SCP4WMS mechanism of the ‘agile-consistent process’, which supports the computing infrastructure of the SOF4WMS and SHM4WMS mechanisms. In order to guarantee the current agile and consistent attributes in the WfMS mechanisms, we serialise the three mechanisms SCP4WMS, SOF4WMS, and SHM4WMS in a sequence list to solve the agile and consistent WfMS problems into three self-autonomous propriety grades: functionalities, non-functionalities, and transactions. Direct feedback is used to adjust the workflow engine online by the TDC, accompanied by three trustworthy propriety grades–measurement, verification, and evaluation—indirectly proposed for the ACM4WMS or ADCV-PN tools by the management and monitoring tool encountered in the LRE or GRE loops. Moreover, this means that, for exception events that cannot be solved by WfMS mechanisms, we transform these with event parameters into the method of the WfMS in order to pursue further solving with LRE or GRE.

In summary, our study is closely related to former works, but it differs from these in terms of two aspects on which we place emphasis simultaneously: self-autonomic computing and trust computing, including the evaluation method [22], which indicates that Equation (13), if applied only to the $CF(R)$ of Equation (9) and not the $CF(R,H)$, has not resolved the problem of incomplete information on the set of $(R,H)$. Similarly, Equations (18) and (19) are all based on the relatively static software test environment while WfMSs are in runtime. Then, we compare the difference between the $P_{STDE}$ and $P_{SEDE}$ by using Equation (18) and compute the $T_{srte}(s)$ by using Equation (19). Indeed, in order to obtain more precise $T_{site}(s)$ and $T_{srte}(s)$, we might update Equations (18) and (19) with the same style as Equation (9) in future works. Furthermore, we plan to implemented the WfMS based on the fundamental features of Internetware [23], and all of these works involve software architecture or service paradigms representing their trustworthiness concentrically, from direct or indirect viewpoints, which also indicate that the $P_{SEDE}$ would be more frequently influenced by the dynamic software test environment while WfMSs are more in runtime than the $P_{STDE}$, which is generated from the relatively static software test environment.

In future work, we plan to conduct simulations or practical industry experiments to verify and evaluate our measurement algorithm for the WfMS mechanism. It is our hope that future works will be completed in the context of the reduction process of WfMS behaviour entropy, given the unavoidable nature of WfMS behaviour entropy increasing.