Krakow, Poland
O c t o b e r  12 - 14, 2009

A b s t r a c t s

Frank Baetke
Global HPC Technology Program Manager Hewlett-Packard / Scalable Computing Infrastructure Organization

HPC, Grids and Clouds: Synergies and Challenges


The TOP500 list and other metrics clearly show that Blade-based architectures are now fully established as the new standard architecture for HPC-Systems. Those systems provide vast opportunities in terms of system efficiency and density but will also lead to new challenges. New node types will provide additional benefits in terms of scalability and usability.

At the software and middleware level Grids have been established successfully in various dimensions and from there a new trend "Cloud Computing" is emerging. Clouds are enabled through automation and virtualization technologies that make it possible to abstract away much of the complexity of accessing vast amounts of computing and storage. Cloud providers can and will pass on the economies of scale of huge data centres through a pay per use model. Current trends, challenges and examples will be discussed and compared to classical Grid environments.

Peter Coveney
University College London Department of Chemistry, UK

Distributed Computing at the Petascale


It now seems likely that PRACE will provide a unified Europe wide distributed environment for production level high end computing. To be maximally effective, this infrastructure will need to be well balanced, supporting not only such massive computing resources but facilitating data movement across fast networks to and from end-users in a secure manner, in conjunction with smart middleware and policies which support various requirements including advance reservations and urgent computing.

These many requirements have contributed to the poor levels of usability of grids, an issue that has concerned us for a number of years. Our own approach uses application virtualization to address grid-based HPC usability. We describe a middleware tool that handles the virtualization of grid applications, called the Application Hosting Environment (AHE). We describe various features of AHE, which provides access to computational grid resources in standard and non-standard ways; it also federates collections of such grids thanks to its interoperability features. New and forthcoming features include support for urgent computing, audited credential delegation for single sign-on based only on local user credentials, and brokering.

We then discuss a few examples of our computational science research that uses 'petascale' resources to achieve scientific results at unprecedented scales and resolution. The applications span a range of domains from investigation of fundamental problems in turbulence through nanomaterials science research to biomedical applications at the forefront of HIV/AIDS and cancer research, as well as cerebrovascular haemodynamics. Much of this work has been performed on TeraGrid's petascale resource, "Ranger", located at Texas Advanced Computing Centre, and on "Intrepid", the IBM Blue Gene/P system at the Argonne Leadership Computing Facility.

William E. Johnston
Senior Scientist, Energy Sciences Network

Progress in Integrating Networks with Service Oriented Architectures / Grids: ESnet's Guaranteed Bandwidth Service


The past several years have seen Grid software mature to the point where it is now at the heart of some of the largest science data analysis systems - notably the CMS and Atlas experiments at the LHC. Systems like these, with their integrated, distributed data management and work flow management routinely treat computing and storage resources as 'services'. That is, resources that that can be discovered, queried as to present and future state, and that can be scheduled with guaranteed capacity. In Grid based systems the network provides the communication among these service-based resources, yet historically the network is a 'best effort' resource offering no guarantees and little state transparency. Recent work in the R&E network community, that is associated with the science community, has made progress toward developing network capabilities that provide service-like characteristics: Guaranteed capacity can be scheduled in advance and transparency for the state of the network from end-to-end. These services have grown out of initial work in the Global Grid Forum's Grid Performance Working Group. The services are defined by standard interfaces and data formats, but may have very different implementations in different networks. An ad hoc international working group has been implementing, testing, and refining these services in order to ensure interoperability among the many network domains involved in science collaborations. This talk will describe these services, their evolution, and their current state.

Domenico Talia
Università della Calabria and ICAR-CNR, Italy

Data Mining and Knowledge Discovery Services in Grids


In the latest years Grids enlarged their horizon as they are going to run business applications supporting consumers and end users. To face those new challenges, Grid environments must support adaptive data management and data analysis applications by offering resources, services, and decentralized data access mechanisms. In particular, according to the service oriented architecture (SOA) model, data mining tasks and knowledge discovery processes can be delivered as services in Grid-based infrastructures. Through a service-based approach we can define integrated services for supporting distributed business intelligence tasks in Grids. Those services can address all the aspects that must be considered in data mining and in knowledge discovery processes such as data selection and transport, data analysis, knowledge models representation and visualization. We worked in this direction for providing Grid based architectures and services for distributed knowledge discovery such as the Knowledge Grid, the Weka4WS toolkit, and mobile Grid services for data mining. This lecture describes a strategy based on the use of Grid services for the design of distributed knowledge discovery services and discuss how Grid frameworks, such those mentioned above, can be developed as a collection of Grid services and how they can be used to develop distributed data analysis tasks and knowledge discovery processes using the SOA model.


ACC Cyfronet AGH

ACC Cyfronet AGH
Institute of Computer Science AGH
Institute of Computer
Science AGH
Institute of Nuclear Physics Polish Academy of Sciences