High Performance Computing High Performance Computing University of Southampton

Friday, 20th April 2001, 16:30 hrs

Title : Technologies for the GRID: A debate.
Speaker : Dr. Denis Nicole and Dr. Simon Cox
High Performance Computing Group (HPC)
University of Southampton
Keywords : GRID Technology, Globus, Nexus, Operating Systems, W3C standards

Abstract :

As an opening piece for the debate we invite you to read the background attached. In the seminar we will debate some of the issues which it raises.


We believe that GRID technology will be based on open standards, not on particular implementations or languages. It is, for instance, already clear to the Globus community that it needed to remove its dependence on the NEXUS communications technology. Other apparently safe technologies also have their problems. The open development model of Linux is at risk from the rapid commercialisation of Linux distributions-key developers are now employees of, for example, RedHat; this company will need to find a way to make a substantial return to shareholders from their efforts. Credible GRID technologies will operate across all operating system platforms, in particular across boundaries between UNIX-like and Windows-based systems. Thus authentication and security mechanisms need to be associated with open technologies, such as SSL, and to be decoupled from UNIX-specific views.

Java is not an open technology; it is now clear that Microsoft will not accept Sun's "ownership" of Java. Microsoft and other large IT concerns will simply not go along with a critical business dependence on a competitor's technology. We have already seen this in, for example, Microsoft's rejection of Adobe font technology in favour of Truetype. There is thus a concerted commercial effort to "kill" Java and the associated advantages it offers to Sun, IBM and Oracle-each one a Microsoft competitor-and replace it with C#.

Some HPC Grid developers are attempting to isolate themselves from Microsoft and the world of commercial and leisure computing. This will be a fatal mistake. Cost-effectiveness is driven by volume and, just a the HPC machine vendors have almost all been forced to turn to Intel for their next-generation architectures, so the engineering and scientific communities will have to adopt the superior software technology which is being constructed for the commercial and leisure market. We are now a long way past the era in which computers were developed to run Atomic weapons codes-processor development is driven by the games requirement and National Defence has to put up with what it gets. In the Scientific community we can see this evolution as we finally move away from X-windows based interfaces-many of which will no longer route through firewalls-and towards using Internet Explorer and Netscape as universal front ends for the open HTML standard.

We believe that the Grid will have to move forward to define its interfaces using open standards that are widely supported in the commercial/leisure community. These either are (such as HTTP and HTML) or soon will be (UDDI, XML Schemas, SOAP) W3C standards. The open marketplace-which includes a complex mix of "free" and "for profit" efforts-will then allow implementations to compete and the best to emerge. We see this playing out now in the continuing competition between Internet Explorer (free, OS linked), Netscape (free, open, portable) and Opera (paid for).


Last updated 06th August 2001. Maintained by M.Molinari.