Efficient Use of Computing Clouds as a Cost-Effective Alternative for High-Performance Computing
Coordinator: Lúcia Maria de Assumpção Drummond
Strategic applications for the country, such as those in the areas of oil, meteorology and biology, depend on high performance computing (High Performance Computing – HPC) to process large amounts of data. Computing clouds have emerged as a low-cost alternative to HPC, offering a set of virtualized resources that can be quickly provisioned and dynamically scalable. However, there are still several barriers to its use. One of them is to make efficient and scalable use of resources and the allocation of virtual machines (VMs), which has a direct impact on performance and financial cost. Furthermore, clouds are prone to failures and to satisfy service level agreements, fault tolerance of VMs in clouds is a major concern.
Another important aspect of clouds in the context of high-performance computing is the provision of Graphics Processing Units (GPUs) in these environments. GPUs have proven to be a powerful and efficient platform for accelerating a large class of compute-intensive applications. For this reason, many large-scale data centers are based on heterogeneous architecture, which includes CPUs and GPUs, to meet high performance and data throughput requirements. Following this trend, GPUs are also being used in computing clouds. Exploiting GPUs in clouds, through GPU virtualization, allows physical devices to be logically decoupled from a compute node and shared by any application, resulting in reduced monetary costs, energy savings, and greater flexibility.
In this scenario, efficient sharing of a GPU across different applications is an indispensable feature. In order to allow better use of GPUs, it is necessary to address the issue of concurrency and kernel scheduling in this environment. In this context, this project addresses issues of resource management, CPUs and GPUs.
Team
| Universidade Federal Fluminense (UFF) |
|---|
Lúcia Maria de Assumpção Drummond   |