In recent years, Oracle has significantly expanded the range of services and products it offers related to cloud computing. Engineered systems like Oracle Exadata are now offered over the cloud. The aim of this project is to test these systems, including when using high-end CPUs and GPUs, in order to assess the feasibility of integrating them into the workflows of the LHC experiments. At the same time, we are also providing Oracle’s engineering teams with feedback that can be used to further improve these products.
When the HL-LHC becomes operational in the mid-late 2020s, the required computing capacity — using current hardware, software, and analysis techniques — is expected to be 50 to 100 times higher than today. It is therefore important to use the most cost-effective computing solutions, which can often entail using resources that become temporarily available in an opportunistic manner. In terms of databases, cloud technology offers a potential disaster-recovery solution, makes it technically feasible to offload read-only workloads and backups to the cloud, and opens the possibility of replacing some private data-centre capacity if deemed advantageous.
In 2017, we worked closely together with multiple LHC experiments and engineering teams from Oracle. Workloads from the LHC experiments were executed using Oracle Cloud Infrastructure resources. We set up infrastructure that makes it possible to run HTCondor jobs inside Docker containers and to integrate cloud resources into the management of workloads for the LHC experiments.
We also focused on functional tests of cloud-provisioned databases, both single-instance databases and Oracle Real Application Clusters. We also worked together with Oracle engineering teams to conduct performance tests comparing engineered systems in the cloud and on-premises databases (including Oracle In-Memory). We gained important insights and shared valuable feedback with the Oracle engineers and product managers related to all areas of work within the project.
We plan to set up a test site for disaster recovery using Oracle Database Cloud services, thus helping us to improve our understanding of the requirements related to this (in terms of personnel, networking, data protection, etc.). We will also evaluate the use of Oracle Exadata to help improve the performance of the most complex workloads. In addition, we intend to repeat the experimental tests carried out this year using the new generation of computing hardware available via the cloud, including making use of HPC resources and machine-learning approaches based on GPUs.