Basic issues about Grid partitioning

groups of resident services are controlled by distinct administrations, which manage the infrastructure among local services and commit the communication infrastructure with the rest of the Grid to third parties.

We argue that, in a Grid that is reasonably engineered, characteristics should be mainly determined by inter-domain fabric.

This conclusion comes from the consideration that single administrations try to exploit as much as possible their share of a common infrastructure: there is little sense in a community that leases a 1Gbps long-haul interconnection, while the internal connectivity is based on a 100 Mbps infrastructure.

In that case only inter-domain characteristics need to be reported, since we assume that intra-domain characteristics are irrelevant as for the task of optimizing the performance of a distributed computation.

We conclude that a one-level hierarchy (let's call it a domain partitioning) may help the task of reporting the characteristics of network services.

However, this solves the complexity problem only if domain sizes tend to grow as rapidly as the numer of domains, since in that case the size of the database is O(n), and if the basic assumption, reported in the paper, holds:

Requirement:
For each pair of domains, the intra-domain communication infrastructure performs better when compared with the inter-domain communication infrastructure.