Figure 1.2 shows each step of the initialization of a simple Grid system. The architecture is built in hierarchical order, each component connecting to its parent. The MA is the first entity to be started (1). It waits for connections from LAs or requests from clients.
In step (2), an LA is launched and registers itself with the MA. At this step of system initialization, two kinds of components can connect to the LA: a SeD (3), which manages some computational resource, or another LA (4), to add a hierarchical level in this branch. When the SeD registers to its parent LA, it submits a list of the services it offers. The agent then reports the new service offering through its parent agent until the MA. If the service was previously unavailable along that arm of the hierarchy the agents update their records. Finally, clients can access the registered service by contacting the MA (5) to get a reference to the best server available and then directly connect to it (6) to launch the computation.
The architecture of the hierarchy is described in configuration files (see Section 10.1.2) and each component transmits the local configuration to its parent. Thus, the system administration can also be hierarchical. For instance, an MA can manage a domain like a university, providing prioritary access to users of this domain. Then each laboratory can run an LA, while each team of the laboratory can run some other LAs to administrate its own servers. This hierarchical administration of the system allows local changes in the configuration without interfering with the whole platform.
