The running theme for SC11 will be data-intensive science, with a large number of presentations and sessions focused on the problems and new developments spawned by “big data” and technical or scientific computing.
According to John Johnson, conference thrust chair and association division director at Pacific Northwest National Laboratory, “Data is a huge challenge in science today…the rapid advancements in data collection and generation are challenging traditional methods of storing, managing and analyzing the information.” He says that this year the supercomputing community is “being called upon to rise to the data challenge and develop methods for dealing with the exponential growth of data and strategies for analyzing and storing large data sets.”
With this theme in mind, we wanted to call your attention to some select sessions and special events at SC11 for those who are exploring data-intensive computing. As Johnson noted, the main issues are analysis, management and storage of large data sets, thus we’ll organize our “must see” elements for the show along those lines with the addition of visualization as another important topic for data-intensive scientific computing.
This year’s emphasis is in line with the announcement of the Graph 500 list, which will be presented on Tuesday afternoon. This will showcase the top of the line systems (according to the benchmark, anyway) for data-intensive computing applications. The organizers hope the session that follows will provide a discussion opportunity that will focus on evolution of the benchmark and the future of data-intensive science. The session can be found here and more information about this notable list can be found at the Graph 500 site.
In advance of the topical breakdown, however, it is worth mentioning that there is a thorough introduction to data intensive computing that runs for the first half of the morning on Monday. This session, presented by Robert Grossman from the University of Chicago and Collin Bennett from the Open Data Group will offer the “big picture” of data intensive computing by touching on utility clouds (Amazon) and data clouds, as provided by Hadoop. They will also provide an introduction to managing scientific datasets using distributed file systems like Hadoop and NoSQL databases like HBase. This will be in addition to parallel programming frameworks, including MapReduce, Hadoop steams and related techniques. It’s a lot to achieve in one short morning but the presents hope to illustrate the role of these and other tools for managing large datasets. If Monday morning is free and you want an initial big data deep dive, this is probably the best session early in the conference.
The Second SC Workshop on Petascale Data Analytics: Challenges and Opportunities workshop, which runs all day on Monday will provide a dense overview on the growth of data intensive applications (and dataset sizes) and show how trends like cloud computing are becoming a way to handle the peak loads and large data demands of emerging applications. This workshop will be hosted by researchers from Oak Ridge National Lab and the University of Minnesota.
Another day-long workshop on Monday focused on data-intensive computing will be presented by Ian Taylor from Cardiff University and Johan Montagnat from CNRS. This event, which is the sixth Workshop on Workflows in Support of Large-Scale Science will focus on the “many facets of data-intensive workflow management systems, ranging from job execution to service management and the coordination of data, service and job dependencies.” The presenters hope to cover a range of related issues throughout the day, including data intensive workflows representation and enactment; designing workflow composition interfaces; workflow mapping techniques that may optimize the execution of the workflow; workflow enactment engines that need to deal with failures in the application and execution environment; and a number of computer science problems related to scientific workflows such as semantic technologies, compiler methods, fault detection and tolerance.
More analysis-related sessions of note include:
Semantic Graph Database Processing
Evaluating NoSQL for Enterprise Applications
Using Semantic Web Technologies on HPC Clouds
There are a number of deeper, specialized sessions on management of big data, but a few do offer some promise for the non-specialist in terms of the tools that are the focus of the session. For instance, Monday’s “Big Data Means Your Metadata Must Work” touches on the range of tools uses to capture and use metadata using real-world examples. The presenters hope to provide attendees with a better sense of the many metadata tools that are available and how can be used to help share big data.
Other management-related sessions to note include:
Parallel Index and Query for Large Scale Data Analysis
Hadoop Acceleration Through Network Levitated Merge
Open source file systems – Transitioning from Petascale to Exascale
I/O Streaming Evaluation of Batch Queries for Data-Intensive Computational Turbulence
With estimates predicting that data growth will surpass Moore’s Law to 1.8 zettabytes by the end of this year and file-based data growing 75 times what it is now over the course of the next decade, the storage piece of the data-intensive computing puzzle is among the most important.
There are a number of presentations during the show, including one from Nick Kirsh of EMC/Isilon, called “Big Data, Big Opportunity: Maximizing the Value of Data in HPC Environments.” Kirsh plans to present “real-life implementations in which scale-out storage dramatically accelerated data and server performance, speeding time-to-results in critical HPC projects to extract maximum value from HPC data.” He also plans to address how implementing scale-storage can work toward eradicating the bottlenecks that HPC users with large datasets encounter.
Other storage-related sessions to note include:
The Sixth Parallel Data Storage Workshop (all-day event)
Terascala – Enabling Fast, Easy to Manage Storage Appliances
One of the highlights this year will be the Scientific Visualization Showcase, which will demonstrate how visualization is being used to model everything from the beginning of the universe to jet engines. While this range of presentations is guaranteed to be great eye candy, there are a relatively large number of visualization presentations this year.
Visualization events outside of the showcase include a workshop on ultrascale visualization presented by Kwan Liu Ma from the University of California, Davis and Michael Papka from Argonne National Lab. The workshop, which runs from 9:00-5:30 on Sunday before the conference kickoff, will address new ways to become capable of exploiting petascale data to its fullest with exascale datasets on the horizon. The two will spend Sunday addressing “the latest and greatest research innovations in large data visualization and how these innovations impact scientific supercomputing and the discovery process.”
A number of the instructional sessions will touch on various elements of large-scale data analysis and visualization, including a tutorial for using the open source visualization and analysis application ParaView, which allows users to visualize large data sets in parallel. Outside of the more application-specific tutorial, the presenters plan to provide more general guidance about visualizing the massive simulations that run on supercomputers—and do a walk through of installation and set-up of ParaView.
Other visualization-related sessions to note include:
Large-Scale Data Visualization for Data-Intensive and High-Dimensional Scientific Data Analysis
World-highest Resolution Global Atmospheric Model and Its Performance on the Earth Simulator
These lists certainly don’t do justice to the wide range of sessions to choose from across the data-intensive computing spectrum and didn’t even begin to touch on the many sessions with clear HPC/big data cross-over appeal. Still, we look forward to see you all in Seattle this year—stop by our booth to share insights you’ve gleaned from these and other presentations, won’t you?