Deliverables

Deliverable D2.4 provides the final version of ACTiCLOUD's cloud resource manager, named as ACTiManager, that consists of functional components, placement algorithms, models for workload classification, interference detection, and performance prediction. While generic in concept and design, ACTiManager is implemented on top of and fully integrated with OpenStack cloud management platform. The deliverable consists of the software source code and its documentation.

This deliverable is tightly coupled with deliverables D1.1: “ACTiCLOUD Requirements and base architecture” and D1.2: “ACTiCLOUD architecture”. It is an extension of deliverable D3.1: “ACTiCLOUD-enabled system libraries v1.0” and strongly linked to the components that are put together in the ACTiCLOUD prototype.

The major achievements of the ACTiCLOUD libraries are the improved memory-allocator NC-ALLOC, and the other NUMA aware libraries (NC-LAPACK, NC-MPI, NC-BLACS, and NC-BTL) that make ACTiCLOUD instances available to Big Data, Analytics and HPC applications that need a lot of memory and low latency. Applications using NC-ALLOC, NC-LAPACK, NC-MPI, NC-BLACS, and NC-BTL show good scaling in benchmarks.

Deliverable D3.6 provides the final version of ACTiCLOUD's custom Java Virtual Machine implementation, named as Hyperscale JVM. The deliverable consists of the software source code and its documentation, with special focus on the tools developed during the project for research on memory management.

This deliverable is a follow-up of D3.3 “MonetDB ACTiCLOUD Extensions v1.0”, in which we have reported the initial implementation of three MonetDB extensions to exploit the functionality enabled by the ACTiCLOUD architecture: i) MALCOM estimates the memory footprint MonetDB needs for a query before its execution; ii) MonetDBLite-Java, an embedded version of MonetDB for Java; and iii) distributed query processing features of MonetDB. In the second period of this project, we have both continued developing the aforementioned software, as well as developing new software. In this deliverable, we present one new MonetDB artefact and give updates to two existing artefacts.

MDBconductor is a new MonetDB artefact. It manages a cluster of MonetDB VM instances in the cloud, and uses the estimation of MALCOM to choose a suitable instance to execute a given query. It also takes care of transparently resizing an instance (e.g. memory and CPU cores) before executing a query, if none of the existing instances is suitable for certain query workload.

MonetDBLite-Java has been ported to the ARMv8 (AArch64) architecture. Furthermore, to ensure we can keep developing and maintaining this MonetDB artefact even after the end of ACTiCLOUD, we have done two major code refactoring. First, we have merged the MonetDBLite and MonetDB codebases into one codebase. MonetDBLite is the core library of MonetDBLite-Java. This merge also enables us to backport features, such as the in-memory only mode, from MonetDBLite to MonetDB. Second, we have migrated the MonetDB build-install-test system from the classic Autotools to the more modern CMake.

To allow MonetDB to scale out to a large number of instances, we have extended MonetDB’s distributed query processing feature to support predicate-based data partitioning and distributed updates.

This document describes the evolution of the Neo4j graph database within the context of theACTiCLOUD project. This deliverable is not independent but relies on Deliverable D3.5:“ACTiCLOUD-enabled system libraries”.

Deliverable D4.4: “ACTiCLOUD Final Prototype” integrates the final (v2.0) versions of the components of the ACTiCLOUD architecture. More specifically, the deliverable focuses on various integration scenarios of the final ACTiCLOUD components, i.e., the two hardware platforms that are available within the project (Numascale and KMAX), the MicroVisor, the OpenStack cloud framework, the ACTiManager, the HyperscaleJVM, the system libraries, and the MonetDB and Neo4j databases. The purpose of the integration scenarios is to demonstrate the interoperability between the major ACTiCLOUD core components. To maximise the impact of the core components, the validity of the integration scenarios, and to solve other challenges that occurred during the course of the project, we also include commodity hardware platforms (Intel and AWS bare metal) and virtualization technologies (KVM and Xen) in our integration scenarios. The deliverable includes documentation together with the corresponding software packages. Finally, the deliverable concludes with the remaining open issues and draws directions for future work.

Deliverable D4.5: "ACTiCLOUD Final Evaluation" describes the evaluation of the final ACTiCLOUDprototype and its constituent components that were described in Deliverable D4.4 "ACTiCLOUDFinal Prototype". The evaluation is based on the strategy and methodology that was described in D4.3 "ACTiCLOUD intermediate evaluation", regarding the strategic objectives of the project and the use cases and business scenarios that the ACTiCLOUD architecture targets.

D5.8 provides the final project press release that highlights the ACTiCLOUD achievements.