Challenges in Data Intensive Computing

Panel Chair: Dr. Rob Ross, Argonne National Laboratory, USA

As computing systems mature, three distinct storage interaction models are emerging.
On desktop systems we are now seeing search capabilities change the way users interact with the file system, lessening the role of the traditional directory hierarchy. In the wide area, storage resource management systems that allow users to locate and retrieve data are becoming an important aspect of the scientific process. On HEC systems the focus remains on extremely high performance, with applications interacting primarily through POSIX and structured I/O interfaces and with interactive access taking a back seat.

Some HEC sites have long pushed for a globally accessible, secure, high-performance parallel file system that stores data from applications, allows direct access to this data from multiple on-site systems, and provides access across the wide area. This vision has persisted for quite some time, despite the inability of the community to make it a reality.
In this panel we will revisit the vision of data access from the user's perspective. We will discuss the viability of this vision from a technical perspective as well as debating how appropriate this vision is given the reality of today's computational science and its workflows.

Questions:

• How would you characterize the models of interaction with HEC storage system?
  • On the HEC machine?
  • At a desktop or other system on-site?
  • Remotely?
• Given these models of interaction, how compelling is the vision of a globally accessible file system?
  • What parts of the vision match with your models of interaction?
  • What is missing from the vision?
  • What is unnecessary?
• What are the biggest technical challenges facing the implementation of your vision?
  • Are there technical challenges that seem insurmountable?
  • What technologies can we adopt from other areas?
  • What should we be focusing on now?