Page 18 - Sirius_Annual_Report_2021

Scalable Computing
The Scalable Computing (SC) research program is about making data access and processing faster
to SIRIUS projects. This is achieved by building knowledge in High Performance Computing (HPC)
and coupling this with scalable Cloud computing to support scalable big-data application processing. Specifically, we look at solutions for scalable and reconfigurable hardware, software design for parallel numerical simulations, and automatic cross-cloud application deployment and reconfiguration using hardware accelerators.
The target problems of HPC involve large scale computa- tions that are beyond the capabilities of single laptop PCs or desktop computers. Moreover, close interactions often exist between the inherent components of these compu- tations, thus the required hardware platforms for HPC
are tightly coupled computer clusters, consisting of many powerful interconnected computers. The research topics
of HPC encompass parallelization schemes, partitioning algorithms, communication overhead reduction strategies, software implementation and optimization techniques, use of heterogeneous clusters that consist of both conventional CPUs and cutting-edge hardware accelerators, in addition to adopting HPC for real-world applications.
The methodology on the HPC side is largely experimentation with different computing platforms built on technology from SIRIUS hardware partners and to evaluate the per- formance of these platforms for real applications of the SIRIUS application partners. The work will therefore involve experimental software design and hardware architectures for scalable computing ranging from accelerators to numerical methods. Stochastic combinatorial optimization is the methodology used for managing applications across different Cloud providers to allocate the application components where they give the best utility for the application owner, and to reconfigure in response to changing application execution context.
The Scalable Computing research have the following objectives:
Objective 1 Better and Flexible Execution Platforms:
The focus is on making advanced execution platforms available to all SIRIUS partners through open interfaces that can be used remotely allowing researchers without direct access to the computers to use and experiment with different hardware configurations for their applications.
Objective 2 Scalable Application Support:
On one side this will continue the support to open source for better numerical computations for reservoir simulations, and on the other side it will continue the development of Cross-Cloud and Multi-Cloud application management middleware.
Objective 3 SIRIUS Application Execution:
To evaluate the research delivered under the previous objectives, demanding real world applications from the SIRIUS partners may be tested using the software and the hardware available.
Activities and their contribution to the main objectives:
Objective 1:
• Work is ongoing to link UiO’s Numascale computer with NREC to enable its use as Cloud HPC platform to demonstrate how HPC applications can benefit from using Cloud computing
• A new and flexible HPC cluster architecture based on PCIe has been established and is under testing.
• Selection has started for the hardware for a future Exascale experimental platform.
Objective 2:
• The research activities on numerical methods for reservoir simulations and the associated code optimizations have continued.
• The work on installing the Cloud management software Open Stack on the moved NUMAScale computer has started.
18 | SIRIUS ANNUAL REPORT 2021