Experiments with a Machine-centric Approach to Realise Distributed Emergent Software Systems

Abstract

Modern distributed systems are exposed to constant changes in their operating environment, leading to high uncertainty. Self-adaptive and self-organising approaches have become a popular solution for runtime reactivity to this uncertainty. However, these approaches use predefined, expertly-crafted policies or models, constructed at design-time, to guide system (re)configuration. They are human-centric, making modelling or policy-writing difficult to scale to increasingly complex systems; and are inflexible in their ability to deal with the unexpected at runtime (e.g. conditions not captured in a policy). We argue for a machine-centric approach to this problem, in which the desired behaviour is autonomously learned and emerges at runtime from a large pool of small alternative components, as a continuous reaction to the observed behaviour of the software and the characteristics of its operating environment. We demonstrate our principles in the context of data-centre software, showing that our approach is able to autonomously coordinate a distributed infrastructure composed of emergent web servers and a load balancer. Our initial results validate our approach, showing autonomous convergence on an optimal configuration, and also highlight the open challenges in providing fully machineled distributed emergent software systems. CCS Concepts •Software and its engineering → Distributed systems organizing principles; Software design engineering;

The State of the Art of Emergent Software Systems

“…Response time and responsiveness Filho et al [8], [9], [12], [18], [19] Response time and responsiveness Cardozo [4] X Wang et al [20]- [22] Responsiveness, availability, throughput, successability, reliability Younes [6] Hybrid (user's feedback) Mannava and Ramesh [23] Process CPU time, heap memory Ding et al [24] X Rosa et al [25] X Tan et al [26] Dependability and responsiveness Gonçalves et al [27] Responsiveness Yan et al [28] Responsiveness Belhaj et al [29] Availability, responsiveness, service calls Schneider et al [30], [31] Throughput, energy costs, efficiency Ganguly and Sakib [32] Failure rate, responsiveness Deshpande et al [33] Response time, availability, throughput, successability Kulkarni et al [34] Response time, model confidence, and CPU consumption Rainford et al [35] Response time, Resource utilization Silva et al [36] Resource utilization…”

“…As seen in Table 3, most ESS approaches are based on nonfunctional adaptation goals, including response time [8], [9], [12], [15]- [19], [33], [35], responsiveness [20]- [22], [27]- [29], availability [20]- [22], [29], throughput [20]- [22], [30], [31], [33], successability [20]- [22], [33], reliability [20]- [22], process CPU time [23], [34], heap memory [23], dependability [26], service calls [29], energy cost [30], [31] and failure rate [32]. Some approaches only target one adaptation goal at a time such as [27], [28], while others tackle multiple adaptation goals.…”

“…Due to its importance, many ESS approaches have been proposed in the literature [6]- [9]. Analyzing the literature of these approaches is important to equip researchers and industrial stakeholders with a comprehensive view of existing ESSs, enabling them to select approaches that meet their needs and helping them identify potential research avenues.…”

Hierarchical Self-Awareness and Authority for Scalable Self-Integrating Systems

“…A specific example is a datacentre software infrastructure (Fig. 3), which receives requests from users and should serve responses as quickly as possible (reduce latency) while using a minimal set of resources (maximise energy efficiency), all while user behaviour is continuously changing to cause stress in different parts of the system [6]. While conceptually simple, in practice this kind of system can involve hundreds of subsystems which are highly challenging even for expert human engineers to design and maintain [23], making it a good target for automated design with runtime learning.…”

“…In large-scale systems operating in complex environments, self-aware controllers must acquire and process increasingly large amounts of knowledge, which can become problematic in terms of delays and resource consumption. Hence, selfintegration controllers are often decentralised [3], [5], [6] ( Fig. 1-b), to parallelise knowledge acquisition and processing (lowering execution times and distributing resource usage).…”