Monitoring a large cloud can mean monitoring potentially thousands of ma-
chines, so operators can not afford to gather and analyze the data in a centralized
fashion. A prototype of a distributed communication and computation system for
monitoring cloud systems has been developed, which is scalable, highly available,
and based on real-time stream processing principles. In essence, it is a publish/-
subscribe system, but when a subscription is made to a transformation of some
data that is not being calculated yet, a new computation process is started that
then provides the desired transformation. With nested transformations, data from
an existing provider is reused, achieving that a certain computation on a piece of
data is only ever performed once in the whole system.
|
