Updates to the text inline. Note that this text should be moved to
be a subsection of identification (2.1.1).
Andrea
What?
Meta-data
attached to one or more attributes from
the information model for a manageable
resource. It indicates that the values of these
"correlatable" attributes can be
compared across instances, and if they are equal, signifies that the same
"real-world" entity is being described by those instances. The use of a
standardized meta-data "label"
provides a non-ambiguous mechanism for
defining attributes appropriate for correlating identities.
It is important to
note the scope within which the attributes are indeed correlatable.
For example, although disk drives have serial numbers, they are found
to repeat. Therefore, within a single
system, the numbers for a single vendor are unique
identifiers. However, across an enterprise or even across multiple
vendor products in a single system, that may not be
true.
Note that the attributes that
are labelled as "correlatable"
are specific to the semantics of the managed resource. For example,
in fibre channel environments, world-wide-names are likely to be globally
unique. Therefore, a single "WWN"
attribute may be labelled with the "globally correlatable" tag for an instance of FCA (fibre channel
adapter). This is independent of how the adpater is discovered or
identified by the management infrastructure.
Why?
The general use case for correlatable names is that a
single "real-world" entity may be managed through several endpoints, and/or managers. For
these endpoints and managers, the instances must be identifiable.
This does not, however, dictate that a single identification scheme must be
adopted throughout the managed environment. To support a single scheme,
all endpoints and managers would need access to the
complete information required by the "identification" algorithm, and it
would be necessary that the algorithm be sufficient in all
implementations.
Since a
globally unique identifier for all environments is not achievable, it is
important to provide mechanisms to determine if data for the same or different "real-world" entities
is being reported. For example, when performing discovery, it is
necessary to accurately determine the resources in the environment (and not
over-estimate). Continuing this
example, "correlatable names" are required when a partitioned system has multiple management
infrastructures - perhaps one for
each of the partitions and one for the host system. The partitions' hardware data will be a strict subset of the host system's data. But, the host system's data will be completely populated, and therefore a superset. There is a need
to correlate these "hardware" instances from multiple
management endpoints, using processor GUIDs and/or other data.
Andrea