GlueDomains: Organization and Accessibility of Network Monitoring Data in a Grid

The availability of the outcome of network monitoring activities,
while valuable for the operation of Grid applications, poses serious
scalability problems: in principle, in a Grid composed of n
resources, we need to keep record of n² end-to-end paths.

We introduce a scalable approach to network monitoring, that consists
in partitioning the Grid into domains, limiting monitoring activity
to the measurement of domain-to-domain connectivity. Partitions must be consistent
with network performance, since we expect that an observed network
performance between domains is representative of the performance between
the Grid Services included into such domains. We argue that partition design
is a critical step: a consequence of an inconsistent partitioning is
the production of invalid characteristics.

The paper discusses such approach, also exploring its limits. We
describe a fully functional prototype which is currently under test in
the frame of the DATATAG project.