The emerging availability of quantum compute resources fosters the examination of its exploitation possibilities in different scientific domains, like artificial intelligence, manufacturing or finance. A significant number of research scientists primarily rely on cloud computing infrastructures while conducting research, whereas access to real (mostly remote) quantum hardware resources requires the deployment and proper configuration of different software components. In this paper we present a hybrid, cloud-based reference architecture that lowers the entry barrier to start new experiments with a wide range of quantum compute resources. The solution facilitates the execution of distributed quantum computing simulations in traditional cloud environments, and also access to several remote quantum compute resources. The reference architecture is highly portable to various cloud platforms, resulting in efficient adaptation and application possibilities by research communities. For demonstration purposes, the paper describes our related experiences on the federated, OpenStack-based research infrastructure of the Hungarian Research Network (abbreviated as HUN-REN).

One of the main driving forces in the era of cyber‐physical systems (CPSs) is the introduction of massive sensor networks (or nowadays various Internet of things solutions as well) into manufacturing processes, connected cars, precision agriculture, and so on. Therefore, large amounts of sensor data have to be ingested at the server side in order to generate and make the “twin digital model” or virtual factory of the existing physical processes for (among others) predictive simulation and scheduling purposes usable. In this paper, we focus on our ultimate goal, a novel software container‐based approach with cloud agnostic orchestration facilities that enable the system operators in the industry to create and manage scalable, virtual IT platforms on‐demand for these two typical major pillars of CPS: (1) server‐side (i.e., back‐end) framework for sensor networks and (2) configurable simulation tool for predicting the behavior of manufacturing systems. The paper discusses the scalability of the applied discrete‐event simulation tool and the layered back‐end framework starting from simple virtual machine‐level to sophisticated multilevel autoscaling use case scenario. The presented achievements and evaluations leverage on (among others) the synergy of the existing EasySim simulator, our new CQueue software container manager, the continuously developed Occopus cloud orchestrator tool, and the latest version of the evolving MiCADO framework for integrating such tools into a unified platform.

Distributed graph neural network training usually starts with an offline partitioner such as METIS, which fixes the layout before any workload is seen. SoftPart GNN is a distributed GraphSAGE runtime on PyTorch DistributedDataParallel (DDP) without one. A short online warm-up turns the vertices that the training sampler touches into an exclusive soft ownership, and a budgeted cache replicates the context that it omits. Training samples only from this local view, so gradient synchronisation is the only per-step cross-worker traffic. A layout-independent full-neighbour evaluator exchanges hidden representations in place of features and, on the four graphs where we verified it, agrees with a single-machine reference to rounding level. We measured a two-layer GraphSAGE of hidden width 16 on OGBN-Arxiv, Reddit, OGBN-Products, IGB-hom-small and IGB-hom-medium at four to 16 nodes with one GPU each, and on four graphs at 32. At four to 16 nodes, SoftPart GNN completes training, placement included, in 22 to 62 percent of the time of a DistDGL baseline on the Deep Graph Library’s (DGL) original distributed sampler, and 21 to 52 percent with the baseline’s partitioning counted. Its test accuracy there stays within 0.006 of DistDGL’s on four of the five graphs. At 32 nodes, the ratio is 0.40 to 0.91. On Reddit, the accuracy is 0.0042 lower, and OGBN-Arxiv is left open. A paired control with modulo ownership and the same cache shows no detected accuracy difference. At four to 16 nodes, the warm-up shortens the training epoch on four graphs and gives no end-to-end gain.