There are three broad classes of information gathered for load balancing:
The configuration of a system (see #tbcnf#2481>)
must be known so it can be compared meaningfully
with other systems. The system state is monitored periodically. The processes
that run may be characterised so that they can be scheduled - this can be done
short term as they run based on very recent history, or long term based over a
number of runs - this is especially true of embedded and real time systems.
#table2482#
Table: Typical System Configuration Information
In addition, in a distributed system, it is necessary to have other
configuration information - in particular the relative proximity of processes,
processors and data. In a large system this may involve interacting with
routing/topological information (e.g. bandwidth and delay/latency and error
rates of lines between systems - for instance FDDI may be faster than bus
access for some systems or in other words remote file access than data
migration to a slower disk).