For John Brooke, a Grid consists of a federation of autonomous MiniGrids. A MiniGrid consists of a heterogeneous collection of computers and other resources, managed by one organisation. A Grid can then be federated with other Grids until we reach "The Grid". This looks a bit like a fractal.

Before organisations will be able to make use of each other's resources, an accounting mechanism has to be in place. This is to be automated in some way, otherwise it will be too difficult for users. The name "Grid" was coined in comparison to the power grid. However, the situation is very different there. There is a unit of power that can be used to specify the resource. In Grid computing, the situation is different. There is for instance the unit of computation "Mflop/s". However, these cannot be added when they come from different resources. For most applications, 1000 times a computer with 1 Mflop/s of performance is not the same as having one processor that performance 1000 Mflop/s.

According to John Brooke, the use of Ticket could provide a solution to this problem. A Ticket is an offer to run a certain job under certain conditions. A ticket is only valid for a limited time.

As an example, a user could request to "Run the DWD local weather model code, over 1000 grid points, simulating over a 24 hour period". A broker could take this request, dispatch it to several organisations and return concretised offers, such as "16 T3E processors for an hour, done by midday, for euro 32", "32 O3000 processors for 30 minutes, done in an hour, for euro 45".

The offers are sent back to the user as a ticket that he can accept and then the job is run, or he can not accept it. The translation of the job requirements to resources that are needed is, in the view of John Brooke, done by the broker.

To let this work, a standard resource description language is needed. Unfortunately, that standard does not yet exist. The resource description language from, for instance Globus, is different from that of Unicore and from that used by commercial resource management systems.

The resource broker under development by the University of Manchester is used in, for instance, the EU funded project EuroGrid. EuroGrid can be considered to be a federation of geographically located centres each of which has MiniGrid structure. The EuroGrid nodes have an independent existence from the Grid project. They predate it and, as John Brooke expects, may well post-date it. This makes the EuroGrid a useful test bed.

An additional problem in Grid computing is that some typical applications require simultaneous coupling of resources. An example are ocean-atmosphere models that are being steered via a visualisation server. According to John Brooke, we need message passing tools liken PACX-MPI and STAMPI that operate across heterogeneous systems.

In the vision of John Brooke, a computational grid does not obviate the need for resource providers. For high-performance computing to work, a high degree of sophistication and architectural differentiation is needed. These resource providers - computer centres - need to benefit from participation in a Grid.

More information on EuroGrid can be found at: http://www.eurogrid.org/. For more information on the other products and projects mentioned, consult: http://EnterTheGrid.com.