An Embryonics Inspired Architecture for Resilient Decentralised Cloud Service Delivery

Data-driven artificial intelligence applications arising from Internet of Things technologies can have
profound wide-reaching societal benefits at the cross-section of the cyber and physical domains. Usecases are expanding rapidly. For example, smart-homes and smart-buildings provide intelligent monitoring, resource optimisation, safety, and security for their inhabitants. Smart cities can manage
transport, waste, energy, and crime on large scales. Whilst smart-manufacturing can autonomously
produce goods through the self-management of factories and logistics. As these use-cases expand further, the requirement to ensure data is processed accurately and timely is ever crucial, as many of these
applications are safety critical. Where loss off life and economic damage is a likely possibility in the
event of system failure. While the typical service delivery paradigm, cloud computing, is strong due
to operating upon economies of scale, their physical proximity to these applications creates network
latency which is incompatible with these safety critical applications. To complicate matters further,
the environments they operate in are becoming increasingly hostile. With resource-constrained and
mobile wireless networking, commonplace. These issues drive the need for new service delivery architectures which operate closer to, or even upon, the network devices, sensors and actuators which
compose these IoT applications at the network edge. These hostile and resource constrained environments require adaptation of traditional cloud service delivery models to these decentralised mobile
and wireless environments. Such architectures need to provide persistent service delivery within the
face of a variety of internal and external changes or: resilient decentralised cloud service delivery.
While the current state of the art proposes numerous techniques to enhance the resilience of services
in this manner, none provide an architecture which is capable of providing data processing services in
a cloud manner which is inherently resilient. Adopting techniques from autonomic computing, whose
characteristics are resilient by nature, this thesis presents a biologically-inspired platform modelled
on embryonics. Embryonic systems have an ability to self-heal and self-organise whilst showing capacity to support decentralised data processing. An initial model for embryonics-inspired resilient
decentralised cloud service delivery is derived according to both the decentralised cloud, and resilience
requirements given for this work. Next, this model is simulated using cellular automata, which illustrate the embryonic concept’s ability to provide self-healing service delivery under varying system
component loss. This highlights optimisation techniques, including: application complexity bounds,
differentiation optimisation, self-healing aggression, and varying system starting conditions. All attributes of which can be adjusted to vary the resilience performance of the system depending upon
different resource capabilities and environmental hostilities.
Next, a proof-of-concept implementation is developed and validated which illustrates the efficacy
of the solution. This proof-of-concept is evaluated on a larger scale where batches of tests highlighted
the different performance criteria and constraints of the system. One key finding was the considerable
quantity of redundant messages produced under successful scenarios which were helpful in terms of
enabling resilience yet could increase network contention. Therefore balancing these attributes are
important according to use-case. Finally, graph-based resilience algorithms were executed across
all tests to understand the structural resilience of the system and whether this enabled suitable
measurements or prediction of the application’s resilience. Interestingly this study highlighted that
although the system was not considered to be structurally resilient, the applications were still being
executed in the face of many continued component failures. This highlighted that the autonomic
embryonic functionality developed was succeeding in executing applications resiliently. Illustrating
that structural and application resilience do not necessarily coincide. Additionally, one graph metric,
assortativity, was highlighted as being predictive of application resilience, although not structural
resilience.