This work presents an event-driven Extract, Transform, and Load (ETL) pipeline serverless architecture and provides an evaluation of its performance over a range of dataflow tasks of varying frequency, velocity, and payload size. We design an experiment while using generated tabular data throughout varying data volumes, event frequencies, and processing power in order to measure: (i) the consistency of pipeline executions; (ii) reliability on data delivery; (iii) maximum payload size per pipeline; and, (iv) economic scalability (cost of chargeable tasks). We run 92 parameterised experiments on a simple AWS architecture, thus avoiding any AWS-enhanced platform features, in order to allow for unbiased assessment of our model’s performance. Our results indicate that our reference architecture can achieve time-consistent data processing of event payloads of more than 100 MB, with a throughput of 750 KB/s across four event frequencies. It is also observed that, although the utilisation of an SQS queue for data transfer enables easy concurrency control and data slicing, it becomes a bottleneck on large sized event payloads. Finally, we develop and discuss a candidate pricing model for our reference architecture usage.

Information privacy is a critical design feature for any exchange system, with privacy-preserving applications requiring, most of the time, the identification and labelling of sensitive information. However, privacy and the concept of “sensitive information” are extremely elusive terms, as they are heavily dependent upon the context they are conveyed in. To accommodate such specificity, we first introduce a taxonomy of four context classes to categorise relationships of terms with their textual surroundings by meaning, interaction, precedence, and preference. We then propose a predictive context-aware model based on a Bidirectional Long Short Term Memory network with Conditional Random Fields (BiLSTM + CRF) to identify and label sensitive information in conversational data (multi-class sensitivity labelling). We train our model on a synthetic annotated dataset of real-world conversational data categorised in 13 sensitivity classes that we derive from the P3P standard. We parameterise and run a series of experiments featuring word and character embeddings and introduce a set of auxiliary features to improve model performance. Our results demonstrate that the BiLSTM + CRF model architecture with BERT embeddings and WordShape features is the most effective (F1 score 96.73%). Evaluation of the model is conducted under both temporal and semantic contexts, achieving a 76.33% F1 score on unseen data and outperforms Google’s Data Loss Prevention (DLP) system on sensitivity labelling tasks.

Protocols in ad hoc networks are not designed with mobility in mind. Recent research reveals that mobility impacts all the layers of the protocol stack. Specifically, more realistic mobility models that are extracted from real user traces for the vehicular and pedestrian scenarios show that wireless nodes tend to cluster around popular locations. The contributions of this paper are two-fold. First, it suggests cross layer design, as a promising approach, in designing ad hoc protocols with mobility in mind. Therefore, it provides a survey of the methodologies used in wireless cross layer studies. Second, it presents a framework for cross layer and flexible ad hoc protocol design, which integrates mobility into protocol design.

Using drone cells to optimize Radio Access Networks is an exemplary way to enhance the capabilities of terrestrial Radio Access Networks. Drones fitted with communication and relay modules can act as drone cells to provide an unobtrusive network connection. The multi-drone-cell placement problem is solved using adapted Dispersive Flies Optimization alongside other meta-heuristic algorithms such as Particle Swarm Optimization and differential evolution. A home-brewed simulator has been used to test the effectiveness of the different implemented algorithms. Specific environment respective parameter tuning has been explored to better highlight the possible advantages of one algorithm over the other in any particular environment. Algorithmic diversity has been explored, leading to several modifications and improvements in the implemented models. The results show that by using tuned parameters, there is a performance uplift in coverage probability when compared to the default meta-heuristic parameters while still remaining within the constraints implied by the problem’s requirements and resource limitation. This paper concludes by offering a study and comparison between multiple meta-heuristic approaches, investigating the impact of parameter tuning as well as analyzing the impact of intermittent restarts for the algorithms’ persistent diversity.