The Grid makes widely distributed computing resources transparently
available to the end-user. As well as purely computational resources,
these include data storage and networking. The European DataGrid
(EDG) collaboration builds software components which enable this
access; the EDG testbeds are the vehicles for testing of the EDG
software.
There are two testbeds: the development testbed and the application
testbed. The application testbed, is intended for semi-production use
by end-users. The development testbed is a key component of the
development, integration, and certification process.
A number of sites (currently around 15 sites in 8 countries) provide
the computational resources for the testbed. Each site provides
services which are typically organized as follows:
- User Interface (UI)
- This machine runs the User Interface software which allows the
end-user to interact with the EDG testbed. This is typically the
machine the end-user logs into to submit jobs to the grid and to
retrieve the output from those jobs.
- Computing Element or Service (CE)
- A computing element consists of one gatekeeper node and
one or more worker nodes. Together these provide computational
resources to the user.
- Gatekeeper (GK)
- This is the frontend of a computing element, accepting jobs,
dispatching them for execution, and returning the output. It provides
a uniform, grid-accessible interface to the computational resources it
manages.
- Worker Node (WN)
- These nodes sit behind a gatekeeper and are
typically managed via a local batch system. The details of this are
hidden from the end-user by the gatekeeper; however, these are the
nodes on which user computations are actually performed.
Consequently, the end-user software is installed on these nodes.
These nodes do not run any EDG daemons,1.1but do have client APIs for accessing EDG services and information.
- Storage Element (SE)
- These nodes provide uniform, high-level access to data storage. The
storage element may control large disk arrays, mass storage systems
and the like; however, the SE interface hides the differences between
these systems allowing uniform user access.
- Monitoring Node (MON)
- The Monitoring Node runs the R-GMA
servlets for the site and ROS, the replica optimization service.
The resources within a testbed site and the total number of sites
change over time as new resources are added to the testbed or are
temporarily withdrawn for reasons such as maintenance.
There are also several nodes which provide shared services and are not
site-specific but shared by various subgroups of the testbed users.
The most visible are the following:
- Resource Broker (RB)
- These machines accept jobs from users (via the User Interface), match
the jobs' requirements to the available testbed resources, and
dispatch the jobs.
- Replica Location Service (RLS)
This machine runs LRC, the local replica catalog (one part of
RLS), and RMC, the replica metadata catalog. The RLS maintains a
database of a Virtual Organization's (see Section 2.3) data
files and associated metadata. These services are used by users and
by grid services to locate appropriate copies of input data files.
- Information Catalog (IC)
- The Information Catalog runs the R-GMA schema and registry
servlets. The registry is where the URLs of all producers and
consumers of information within the grid can be found. Both users and
grid services use this extensively in the course of normal operation.
There are numerous other services which support, for example, the EDG
security model, but which are used only indirectly by end-users.
http://marianne.in2p3.fr