Content Placement Strategies for Smart Products

Driven by advances in technology as well as standardisation efforts, the adoption of smart things gathers pace in various industries. Smart things range from simple objects being equipped with smart labels to comprehensive objects with embedded storage, computing, and networking capabilities. They aim at improving product operation and usage as well as at enabling functionality beyond their original purpose of use. Think of connected cars that seek for enhancing traffic safety and efficiency as well as for enabling telematic services, or industrial assets that sense and monitor their condition in order to improve maintenance operations. This thesis addresses smart products, a specific class of smart things that targets simplicity of product use by means of multi-modal and personalised product-to-user interaction. For this purpose, smart products are equipped with knowledge and knowledge-related functionality. Amongst others, this includes domain and problem-solving knowledge that is used to enable active user guidance. Given their resource limitation and potential intermittent connectivity, smart products are typically not able to locally store all content required and created across their life cycle neither can they connect to remote systems at all times for storing and / or retrieving information. Yet, in order to achieve simplicity of product use, it is essential that product-to-user interaction is not distorted by long user-perceived delays for retrieving required information or by falling back to simple interaction means, because of content being not accessible. Hence, there is a need for intelligent mechanisms that make content available when required and accessible with low latency. This thesis studies how procedural problem-solving knowledge associated with smart products can be utilised by content placement strategies, i.e., content replication and replacement strategies, to enhance content access. According to the results of the EU-funded research project SmartProducts, the thesis assumes procedural problem-solving knowledge to be modelled in the form of workflows that reflect sequences of activities and feature annotation of activity-related content needs. Also, it assumes complex-structured workflows that consist of multiple alternative branches with the actual branch(es) to be followed being determined dynamically. On this basis, the thesis contributes three workflow-based replication strategies, namely Most Probable Path (MPP), Path Assessment (PA), and Cooperative Path Assessment (CPA). These strategies utilise workflow structures to predict and pre-replicate (i.e., replicate in advance) workflow-related content needs in order to enhance query efficiency of content requests during workflow execution. While MPP, PA, and CPA pursue the same objective, they vary in the way of balancing enhancement of query efficiency with the potential risk of false pre-replication as well as in their incorporated sphere of knowledge. In order to address uncertainty in execution of branched workflows and the resulting risk of falsely pre-replicated content, the workflow-based replication strategies apply the proposed concept of transient replicas. Transient replicas are assigned gradually increasing Time To Live (TTL) intervals. While they may be persisted in case the number of requests exceeds a pre-defined threshold, they are removed in case they are not accessed within their active TTL. This avoids pre-replication wastage in the long term and limits the number of replicas to be maintained. Moreover, the replication strategy Content Class (CC) is proposed. Instead of relying on observed or predicted access patterns, this approach enables smart product applications to classify their intended content use and provides content-class-specific best effort policies to improve content organisation. Given the typical benefits of integrated content replication and replacement strategies, this thesis proposes the two replacement strategies MFR for Transient Replicas (MFRTR) and Enhanced Content Replacement (ECR) that complement the workflow-based replication strategies and CC, respectively. MFRTR extends the well-known replacement policy MFR [KRT06, KRT07] by taking into account transient replicas with lazy removal properties. Thus, transient replicas are only removed if additional storage capacity is needed for serving pending storage operations. This results in efficient utilisation of available storage capacity without increasing replacement overhead. ECR extends MFRTR by leveraging content classification for determining replacement candidates. Instead of purely relying on access frequency and recency, this approach aims at preserving location properties of content according to its classification. The value of the proposed content placement strategies is assessed with a simulation-based evaluation study that adopts the application scenario smart aircraft manufacturing developed jointly with EADS Innovation Works. This study reveals that the proposed strategies improve overall and workflow-related query efficiency by up to 28% and 62%, respectively, compared to related work. As to the knowledge of the author, the proposed content placement strategies are the first that leverage procedural problem-solving knowledge associated with smart products and enable application-affected content organisation while taking into account the challenges of smart products.