SInHPC2 2011 - SInHPC2: Workshop on Software Engineering Innovations for HPC Clouds

SInHPC2: Workshop on Software Engineering Innovations for HPC Clouds

May 20, 2011, Anchorage, Alaska, USA

in conjunction with

International Parallel and Distributed Processing Symposium - IPDPS 2011
May 16-20, 2011, Anchorage, Alaska, USA

Introduction
Large-scale computing systems comprising thousands of computing resources are an essential part of our information infrastructure, as exemplified by the emerging cloud computing paradigm. These systems are the workhorses that drive scientific discovery and enterprise activities by routinely supporting advanced, mission-critical applications in a myriad of fields. Enterprise applications (e.g., economics modeling, keyword search, market-data analysis, social networking, etc.), large-scale high-energy physics experiments, drug discovery, understanding climate change, all are examples of applications that require high performance, reliability, and availability, within limited budgets. Meeting these advanced demands requires innovation in engineering the High Performance Computing (HPC) software that runs these applications.

Current architectural trends pose unprecedented challenges to the HPC programmer. Many-core architectures and heterogeneous compute engines (e.g., GPUs, IBM Cell processor, FPGAs, tiled architectures) make it possible to derive maximum performance by leveraging on-chip and off-chip parallelism. But this performance often comes at a price of having to deal with resource heterogeneity and asymmetry, thus jeopardizing programmer productivity. Innovation in tools and techniques is required to integrate such resources with the cloud computing paradigm, leading to HPC clouds, a model that provides a robust, flexible paradigm for supporting high-level management and execution of large-scale enterprise/scientific applications. Additional benefits include on-demand automatic scaling to a large 2number of compute, storage and networking resources.

The raison d'être of HPC is to reduce time-to-discovery, a metric defined as the total time elapsing between posing a problem and arriving at a solution. Time-to-discovery is largely the sum of the time to develop an HPC application and the time to run it (i.e., Timediscovery = Timedevelop + Timerun). Although arithmetic suggests that reducing either Timedevelop or Timerun will in turn reduce Timediscovery, traditionally HPC researchers have focused on reducing Timerun. The goal of this workshop is to focus on the other summand of this central equation of HPC research. Indeed, Timedevelop constitutes a significant time expenditure for the HPC programmer and can be reduced significantly by applying solid software engineering principles, novel methodologies, and advanced software construction tools.

State-of-the-art software engineering is starting to make inroads into HPC. A major challenge is that HPC applications are often written by domain experts ? scientists and engineers ? who are extremely knowledgeable in their respective domains but may lack a deep understanding of computing or experience with modern developments in software engineering. This problem is compounded by the rapid turnaround in HPC architectures, which makes it impossible for domain experts to become experts in the performance characteristics of any one particular HPC system.

Recent changes in the computing landscape place the issues of Software Engineering in the forefront of the HPC research agenda. The rise of cloud computing opens up unprecedented opportunities to flexibly leverage abundant computing resources on demand, in a scalable and cost-effective fashion. However, for HPC applications to take advantage of these opportunities requires that we fundamentally reconsider how we develop and maintain HPC applications. Software Engineering innovation is required to ensure that the HPC community can reap tangible benefits as users of cloud computing.

To that end, SInHPC2, the Software Innovation in HPC Clouds workshop, is being organized to fill the need for an inter-disciplinary workshop that will bring together researchers and practitioners from the Software Engineering and High Performance Computing communities to explore the challenges, opportunities, and approaches of cloud resources into the service of HPC.