Engineering a Highly Scalable Object-Aware Process Management Engine Using Distributed Microservices

Abstract

Scalability of information systems has been a research topic for many years and is as relevant as ever with the dramatic increases in digitization of business processes and data. This also applies to processaware information systems, most of which are currently incapable of scaling horizontally, i.e., over multiple servers. This paper presents the design science artifact that resulted from engineering a highly scalable process management system relying on the object-aware process management paradigm. The latter allows for distributed process execution by conceptually encapsulating process logic and data into multiple interacting objects that may be processed concurrently. These objects, in turn, are represented by individual microservices at run-time, which can be hosted transparently across entire server clusters. We present measurement data that evaluates the scalability of the artifact on a compute cluster, demonstrating that the current prototypical implementation of the run-time engine can handle very large numbers of users and process instances concurrently in single-case mechanism experiments with large amounts of simulated user input. Finally, the development of scalable process execution engines will further the continued maturation of the data-centric business process management field. 1 Problem Definition For decades, researchers have been examining parallelism, concurrency, and scalability in computer hard-and software. The topic of scalability also became instantly relevant to workflow management systems (WfMS) when they first showed up on the market, as they were explicitly built with large-scale applications in mind [16]. First attempts to create scalable WfMS applied existing scalable architecture principles. The resulting approaches, such as WIDE [7], OSIRIS [17], ADEPTdistribution [5], and Gridflow [6], focused on the system architecture point of view, largely ignoring other aspects, such as role assignments, permissions, and data flow. However, the process models these approaches, especially Gridflow, are meant to support, are typically high-performance production workflows, where these aspects merely play a secondary role [16].

ViePEP-C: A Container-Based Elastic Process Platform

“…In [53], the authors present a highly scalable objectaware process management engine that is using distributed microservices on cloud resources. As an underlying architecture, the authors are using the actor model, where each microservice is an actor (comparable to our services) and connected to different other actors according to a data model.…”

Coordinating large distributed relational process structures

“…Moreover, for object-aware process management, distributed process coordination is of particular importance, as the run-time engine of object-aware process management has a hyperscale architecture [2]. The term hyperscale denotes the ability of the process engine to effectively scale with additionally provisioned resources to provide more performance when computing demands are increasing.…”

“…First, it is shown that the decentralization of coordination constraints over multiple coordination processes not only has conceptual benefits for modelers. The decentralization also enables significant performance increases in a hyperscale architecture [2]. A corresponding experiment with the goal of showcasing this performance advantage has been performed.…”

“…The distribution of coordinators has many advantages, while at the same time only small costs incur [2,7,36]. Adding more decentralized coordinators may still yield more benefits [35].…”

“…When in state Sent, Reviews may be created for the Application (cf. Coordination Constraint 2), a constraint that is represented again by a top-down semantic relationship 2 . Multiple lower-level processes (Reviews) depend upon the execution status (state Sent) of one higherlevel process (the Application) (cf.…”