Shane Canon

Computer centers such as NERSC and OLCF have traditionally focused on delivering computational capability that enables breakthrough innovation in a wide range of science domains. Accessing that computational power has required services and tools to move the data from input and output to computation and storage. A pivot to data is occurring in HPC. Data transfer tools and services that were previously peripheral are becoming integral to scientific workflows.  Emerging requirements from high-bandwidth detectors, highthroughput screening techniques, highly concurrent simulations, increased focus on uncertainty quantification, and an emerging open-data policy posture toward published research are among the data-drivers shaping the networks, file systems, databases, and overall HPC environment. In this paper we explain the pivot to data in HPC through user requirements and the changing resources provided by HPC with particular focus on data movement. For WAN data transfers we present the results of a study of network performance between centers

Flash storage and other solid-state storage technolo-gies are increasingly being considered as a way to address the growing gap between computation and I/O. Flash storage has a number of benefits such as good random read performance and lower power consumption. However, it has a number of challenges too, such as high cost and high-overhead for write operations. There are a number of ways Flash can be integrated into HPC systems. This paper will discuss some of the approaches and show early results for a Flash file system mounted on a Cray XE-6 using high-performance PCI-e based cards. We also discuss some of the gaps and challenges in integrating flash intoHPC systems and potential mitigations as well as new solid state storage technologies and their likely role in the future

Emerging sensor networks, more capable instruments, and ever increasing simulation scales are generating data at a rate that exceeds our ability to effectively manage, curate, analyze, and share it. Data-intensive computing is expected to revolutionize the next-generation software stack. Hadoop, an open source implementation of the MapReduce model provides a way for large data volumes to be seamlessly processed through use of large commodity computers. The inherent parallelization, synchronization and fault-tolerance the model offers, makes it ideal for highly-parallel data-intensive applications. MapReduce and Hadoop have traditionally been used for web data processing and only recently been used for scientific applications. There is a limited understanding on the performance characteristics that scientific data intensive applications can obtain from MapReduce and Hadoop. Thus, it is important to evaluate Hadoop specifically for data-intensive scientific operations -- filter, merge and reorder-- to understand its various design considerations and performance trade-offs. In this paper, we evaluate Hadoop for these data operations in the context of High Performance Computing (HPC) environments to understand the impact of the file system, network and programming modes on performance.

The Cray XE architecture has been optimized to support tightly coupled MPI applications, but there is an in- creasing need to run more diverse workloads in the scientific and technical computing domains. These needs are being driven by trends such as the increasing need to process “Big Data”. In the scientific arena, this is exemplified by the need to analyze data from instruments ranging from sequencers, telescopes, and X-ray light sources. These workloads are typically throughput oriented and often involve complex task dependencies. Can platforms like the Cray XE line play a role here? In this paper, we will describe tools we have developed to support high-throughput workloads and data intensive applications on NERSC’s Hopper system. These tools include a custom task farmer framework, tools to create virtual private clusters on the Cray, and using Cray’s Cluster Compatibility Mode (CCM) to support more diverse workloads. In addition, we will describe our experience with running Hadoop, a popular open-source implementation of MapReduce, on Cray systems. We will present our experiences with this work including successes and challenges. Finally, we will discuss future directions and how the Cray platforms could be further enhanced to support these class of workloads.

We compare the I/O performance using IOR benchmarks on two cloud computing platforms - Amazon and the Magellan cloud testbed.

In this paper we detail benchmarking results that characterize the virtualization overhead and its impact on performance. We also examine the performance of various interconnect technologies with a view to understanding the performance impacts of various choices. Our results show that virtualization can have a significant impact upon performance, with at least a 60% performance penalty. We also show that less capable interconnect technologies can have a significant impact upon performance of typical HPC applications. We also evaluate the performance of the Amazon Cluster compute instance and show that it performs approximately equivalently to a 10G Ethernet cluster at low core counts.