Last week at the ISC Cloud event in Frankfurt, Germany, a sizable group of well-known figures in both HPC and cloud gathered to discuss the progress, challenges and issues on the horizon for high-performance computing in the cloud.
The event, while only in its first year, marks a significant milestone in the HPC cloud space as it shows that there is interest building at the intersection between clouds and technical computing. The sessions provided rich fodder for those interested in understanding the way cloud can enhance current projects or, at the very least, bring about some cost-efficiency.
Wolfgang Gentzsch, ISC Cloud ’10’s chair and expert on grids and the carryover to clouds, stated at the conference opening that, “The high-performance computing community has been deluged with news about the new computing paradigm of cloud computing. Although there is a lot of discussion, there are also many unanswered questions.” While the conference sought to address or at least touch on the vast range of questions that are floating around both inside and outside of the HPC community, among some of the more pressing was whether or not clouds will replace supercomputers.
In Gentzsch’s view, “Many people still seem to be confused about high performance computing (HPC) versus grids versus clouds, and are unsure about the next steps. For ISC Cloud ’10 we are building a program that will answer these questions and help attendees see the cloud more clearly.”
The conference did indeed address many of the most persistent questions about clouds for high-performance computing applications. For instance, Phillippe Massonet from CETIC presented on the topic of “Security in the Cloud: Benefits, Risks and Recommendations” and analysts, including John Barr of the 451 Group, were able to provide some context for the discussions with facts and figures about current adoption, benefits and challenges of cloud for HPC.
Highlights from the Lineup
The list of sessions included discussions ranging from the basic and introductory to some that examined technical issues involved with large-scale cloud deployments. Furthermore, the event drew notable speakers from some of the world’s leading cloud computing and HPC companies and research centers — a fact that provided the first-year conference some name power to draw in attendees.
Dan Reed, Microsoft’s corporate vice president, technology strategy and policy and extreme computing group, presented his keynote, entitled “Technical Clouds: Seeding Discovery,” which hinged on his view that the burgeoning cloud infrastructure is “far bigger than anything previously contemplated in high-performance computing.” Reed’s talk drove home the point that cloud architectures and software paradigms differ from those used in technical computing, and that understanding this is a key first step in approaching HPC clouds, from both an end user and abstract point of view.
Before the event, Wolfgang Gentzsch interviewed Dan Reed about his views on the cloud for scientific computing, which can be read here.
Another leader in the HPC and cloud space who presented a keynote in Frankfurt was Kathy Yelick, associate laboratory director for Computing Sciences and director of the NERSC Division at Lawrence Berkeley National Lab. Her discussion, entitled “Science in the Clouds: A View from Berkeley,” presented some results from the cloud computing testbed for scientific computing at the National Energy Research Scientific Computing Center (NERSC). Berkeley has been a hotbed of cloud computing research, particularly in terms of its pioneering work with both public and private clouds (AWS and its Magellan cloud respectively).
Dr. Ulrich Schwickerath, project manager at CERN, gave an in-depth overview of CERN’s cloud computing infrastructure in which he detailed the research center’s IaaS setup and its successes. More specifically, he discussed the concepts underlying virtualized batch systems and the center’s plans for a virtual batch farm extention.
As Schwickerath explained, “In Spring 201, about 500 recent batch worker nodes have been added temporarily to the system, which allowed us to perform large-scale tests of the new infrastructure. The batch computing farm, which makes up a critical part of the CERN datacenter, now had the possibility to use this IaaS to provision a large number of virtual batch worker nodes. By making use of the new equipment, both the virtual machine provisioning systems and the batch application itself have been tested extensively at large scale. This way it has been demonstrated that the system can sustain 15,000 or more concurrent virtual batch worker nodes.”
Industry Perspectives on HPC Clouds
While scientific computing in the cloud was at the heart of several talks in Frankfurt, a number of industry experts and those with perspectives on cloud adoption in the enterprise also provided context for discussion.
Boyd Davis, vice president of Intel’s Architecture Group, presented “Enabling Scalable and Secure Cloud Infrastructures to Meet the demands of Compute-Intensive Workloads” in which he discussed a number of issues outside of the title, including cloud security, power efficiency, and making choices about scalable storage. The focus of the presentation was on providing a view of Intel’s roadmap of capabilities that are required before wider cloud adoption occurs, and to this end Boyd Davis provided information about available reference architectures in the context of customer examples, particularly those in compute-intensive environments.
Fritz Ferstl, director of grid and cluster management for Oracle Germany, presented use cases for production clouds across industry in order to show how clouds are being adopted by different markets and how these differing segments have made use of them. This is one of the more helpful conversations to have at a cloud-related event because oftentimes, when discussed in the abstract, the challenges of practical cloud implementation are glossed over or not addressed in a balanced way. One of the main questions that Ferstl addressed focused on how industry chooses its cloud solutions, public, private or hybrid, and for those who have not yet made the leap to clouds, what might be preventing the move.
Back for More?
The event is a spin-off conference put together by the same group that brings us the International Supercomputing Conference (ISC) each year, which is organized by Professor Hans Meuer (pictured right) and his Prometeus Team. While there was some talk of clouds at this past year’s ISC conference, the focus remained on traditional, non-virtualized HPC. Having a separate conference dedicated to exploring HPC and clouds allows for greater focus on particular issues that might have been brushed over during an event that was not cloud-centered.
Judging by the response garnered from a first-year event, there is enough momentum gathering in the HPC cloud space to justify its own set of sessions — enough so that it was able to pull in some big names from both high-performance computing and cloud.