Optimising AI Training Deployments using Graph Compilers and Containers
Artificial Intelligence (AI) applications based on Deep Neural Networks (DNN) or Deep Learning (DL) have become popular due to their success in solving problems likeimage analysis and speech recognition. Training a DNN is computationally intensive and High Performance Computing(HPC) has been a key driver in AI growth. Virtualisation and container technology have led to the convergence of cloud and HPC infrastructure. These infrastructures with diverse hardware increase the complexity of deploying and optimising AI training workloads.DeepIaC: Deep Learning-based Linguistic Anti-Pattern Detection for Infrastructure-as-Code
Deep Leaning and code embedding based approach to linguistic detecting anti-patterns in Infrastructure as code. This is from SODALITE smell and defect prediction task.COCOS: a Scalable Architecture for Containerized Heterogeneous Systems
Journal of 2020 IEEE International Conference on Software Architecture (ICSA) pp. 103-113 The work presents a scalable architecture for controlling heterogenous systems. The architecture exploits containers and provides multiple levels of control (container, VM, cluster). A prototype based on Kubernetes is also presented and evaluated.Towards Semantic Detection of Smells in Cloud Infrastructure Code
10th International Conference on Web Intelligence, Mining and Semantics (WIMS) This paper presents our (SODALITE) knowledge-driven approach enabling developers to identify the smells in deployment descriptions /infrastructure codes. We detect smells with SPARQL-based rules over pattern-based OWL 2 knowledge graphs capturing deployment models.SDSN@RT: A middleware environment for single-instance multi tenant cloud application
Journal of Software: Practice and Experience Volume49, Issue5, May 2019 Pages 813-839 The work presented in the mentioned paper reflects the state of art in multi-tenant cloud applications and microservices. It supported the preparation of the SODALITE proposal. The deployment refactoring task in the SODALITE adopts and extends the findings of the paper. Moreover, SODALITE case studies use microservice-based architectures. As the paper is a journal paper, no travel expenses have been claimed.FM4SN: A Feature-Oriented Approach to Tenant-Driven Customization of Multi-Tenant Service Networks
2019 IEEE International Conference on Services Computing (SCC) in Milan (Italy) The presented work includes a novel approach to customize and adapt (micro) service networks. The deployment refactoring support in the SODALITE adopts and extends this approach. Moreover, the two SODALITE use cases use microservice-based architectures.AIOps for a Cloud Object Storage Service
2019 IEEE International Congress on Big Data (BigDataCongress) in Milan (Italy). In this paper the authors share their experience applying the AIOps approach to a production cloud object storage service to get actionable insights into systems' behaviour and health. A real-life production cloud scale service and its operational data are described, the developed AIOps platform is presented, and it is shown how it has helped in resolving operational pain points.A Case for Data Centre Traffic Management on Software Programmable Ethernet Switches
IEEE International Conference on Cloud Networking, 4-6 November 2019 in Coimbra (Portugal). This paper researched the issue with the vendor-independent programmability of traffic and buffer management of virtual switches in the cloud infrastructure, and proposed a way of how to implement this programmability with software-defined networking technologies, in particular with P4.HyperSpark: A Data-Intensive Programming Environment for Parallel Metaheuristics.
2019 IEEE International Congress on Big Data (BigDataCongress), pp. 85-92. IEEE, 2019 in Milan (Italy). The paper presents HyperSpark, which is an optimization framework for the scalable execution of user-defined, computationally-intensive heuristics. The deployment refactoring and optimization support of SODALITE adopts and extends some of meta-heuristics in HyperSpark.