Remote Graphics Processing Units Research Articles

Exploring the use of data compression for accelerating machine learning in the edge with remote virtual graphics processing units

SummaryInternet of Things (IoT) devices are usually low performance nodes connected by low bandwidth networks. To improve performance in such scenarios, some computations could be done at the edge of the network. However, edge devices may not have enough computing power to accelerate applications such as the popular machine learning ones. Using remote virtual graphics processing units (GPUs) can address this concern by accelerating applications leveraging a GPU installed in a remote device. However, this requires exchanging data with the remote GPU across the slow network. To address the problem with the slow network, the data to be exchanged with the remote GPU could be compressed. In this article, we explore the suitability of using data compression in the context of remote GPU virtualization frameworks in edge scenarios executing machine learning applications. We use popular machine learning applications to carry out such exploration. After characterizing the GPU data transfers of these applications, we analyze the usage of existing compression libraries for compressing those data transfers to/from the remote GPU. Our exploration shows that transferring compressed data becomes more beneficial as networks get slower, reducing transfer time by up to 10 times. Our analysis also reveals that efficient integration of compression into remote GPU virtualization frameworks is strongly required.

Redesigning the rCUDA communication layer for a better adaptation to the underlying hardware

SummaryThe use of Graphics Processing Units (GPUs) has become a very popular way to accelerate the execution of many applications. However, GPUs are not exempt from side effects. For instance, GPUs are expensive devices which additionally consume a non‐negligible amount of energy even when they are not performing any computation. Furthermore, most applications present low GPU utilization. To address these concerns, the use of GPU virtualization has been proposed. In particular, remote GPU virtualization is a promising technology that allows applications to transparently leverage GPUs installed in any node of the cluster. In this paper, the remote GPU virtualization mechanism is comparatively analyzed across three different generations of GPUs. The first contribution of this study is an analysis about how the performance of the remote GPU virtualization technique is impacted by the underlying hardware. To that end, the Tesla K20, Tesla K40, and Tesla P100 GPUs along with FDR and EDR InfiniBand fabrics are used in the study. The analysis is performed in the context of the rCUDA middleware. It is clearly shown that the GPU virtualization middleware requires a comprehensive design of its communication layer, which should be perfectly adapted to every hardware generation in order to avoid a reduction in performance. This is precisely the second contribution of this work, ie, redesigning the rCUDA communication layer in order to improve the management of the underlying hardware. Results show that it is possible to improve bandwidth up to 29.43%, which translates into up to 4.81% average less execution time in the performance of the analyzed applications.

Enhancing large-scale docking simulation on heterogeneous systems: An MPI vs rCUDA study

Virtual Screening (VS) methods can considerably aid clinical research by predicting how ligands interact with pharmacological targets, thus accelerating the slow and critical process of finding new drugs. VS methods screen large databases of chemical compounds to find a candidate that interacts with a given target. The computational requirements of VS models, along with the size of the databases, containing up to millions of biological macromolecular structures, means computer clusters are a must. However, programming current clusters of computers is no easy task, as they have become heterogeneous and distributed systems where various programming models need to be used together to fully leverage their resources. This paper evaluates several strategies to provide peak performance to a GPU-based molecular docking application called METADOCK in heterogeneous clusters of computers based on CPU and NVIDIA Graphics Processing Units (GPUs). Our developments start with an OpenMP, MPI and CUDA METADOCK version as a baseline case of cluster utilization. Next, we explore the virtualized GPUs provided by the rCUDA framework in order to facilitate the programming process. rCUDA allows us to use remote GPUs, i.e. installed in other nodes of the cluster, as if they were installed in the local node, so enabling access to them using only OpenMP and CUDA. Finally, several load balancing strategies are analyzed in a search to enhance performance. Our results reveal that the use of middleware like rCUDA is a convincing alternative to leveraging heterogeneous clusters, as it offers even better performance than traditional approaches and also makes it easier to program these emerging clusters.

Multi-tenant virtual GPUs for optimising performance of a financial risk application

Graphics Processing Units (GPUs) are becoming popular accelerators in modern High-Performance Computing (HPC) clusters. Installing GPUs on each node of the cluster is not efficient resulting in high costs and power consumption as well as underutilisation of the accelerator. The research reported in this paper is motivated towards the use of few physical GPUs by providing cluster nodes access to remote GPUs on-demand for a financial risk application. We hypothesise that sharing GPUs between several nodes, referred to as multi-tenancy, reduces the execution time and energy consumed by an application. Two data transfer modes between the CPU and the GPUs, namely concurrent and sequential, are explored. The key result from the experiments is that multi-tenancy with few physical GPUs using sequential data transfers lowers the execution time and the energy consumed, thereby improving the overall performance of the application.

A Resource Selection System for Cycle Stealing in GPU Grids

This paper presents a resource selection system for exploiting graphics processing units (GPUs) as general-purpose computational resources in desktop Grid environments. Our system allows Grid users to share remote GPUs, which are traditionally dedicated to local users who directly see the display output. The key contribution of the paper is to develop this novel system for non-dedicated environments. We first show criteria for defining idle GPUs from the Grid users’ point of view. Based on these criteria, our system uses a screensaver approach with some sensors that detect idle resources at a low overhead. The idea for this lower overhead is to avoid GPU intervention during resource monitoring. Detected idle GPUs are then selected according to a matchmaking service, making the system adaptive to the rapid advance of GPU architecture. Though the system itself is not yet interoperable with current desktop Grid systems, our idea can be applied to screensaver-based systems such as BOINC. We evaluate the system using Windows PCs with three generations of nVIDIA GPUs. The experimental results show that our system achieves a low overhead of at most 267 ms, minimizing interference to local users while maximizing the performance delivered to Grid users. Some case studies are also performed in an office environment to demonstrate the effectiveness of the system in terms of the amount of detected idle time.