|
Dirk Riehle, riehle@acm.org, www.riehle.org
Credit Suisse, Postfach 100, 8070 Zurich, Switzerland.
Position paper for OOPSLA '99 Workshop on Metadata and Active Object Models.
Paper location: http://www.riehle.org/papers/1999/oopsla-1999-ws-21-pp.html
1 Definition
Metadata are data about (some other) data. Metadata describe
the structure and meaning of this other data. Metadata control
processes and processing steps of this other data. Other data
means any kind of data, be it base data or metadata.
2 Examples
2.1 RDBMS example
Consider an RDBMS table description of a CUSTOMER table. The
table description defines that the customer table has columns
like CUSTOMER_ID, FIRST_NAME, LAST_NAME, etc. This table description
is data, and it is different data than a row in the table representing
a specific customer.
The table description is an example of metadata. A row in this
table is base data.
2.2 Business rules example
Consider the data extraction process of loading data from a
host-based operational system onto a data warehousing database.
Customer data flows from the host to the data warehousing database.
During the process, transformation rules check the customer data
for quality and make certain changes to the data.
The transformation rules for cleansing the data are an example
of metadata. The customer data is base data.
2.3 Object-oriented example
Consider an object-oriented system of banking products. A key
product category is accounts. Typically, there are many hundreds
of different account types. Therefore, the system has a class
Account and a class AccountType that provides information about
a specific type of account.
The class AccountType provides metadata about specific Account,
which are its base data. (By the way, AccountType is an instance
of the Type Object pattern.)
3 Classification
Metadata may be
- active or passive;
- business or technical;
- about other metadata or about base data.
3.1 Active and passive metadata
Active metadata is metadata that controls its base data and
is operational in a general sense of the word. It not only describes
its base data but defines how the base data is to be interpreted
and used. In object-oriented systems, metadata become metaobjects,
which unite the metadata with the associated control functions
of the base data.
Passive metadata is purely descriptive metadata that is not
used to control anything. Examples of passive metadata are textual
descriptions like comments of some base data. Passive metadata
is directed at the end-user who knows how to read and interpret
it.
3.2 Business or technical metadata
Metadata has different users. Depending on these users, metadata
may either be viewed as business metadata, describing business
concepts and items, or technical metadata, describing technical
concepts or items. Technical concepts may comprise how business
data is mapped on to an implementation structure, for example,
how a high-level E/R model is mapped onto an RDBMS schema.
Frequently, business data directly maps on technical data,
so that there is no need to distinguish these two categories.
Also, this classification may be extended to incorporate further
stakeholders of the data warehousing process like administrators.
3.3 Meta-metadata
The high road of metadata is data about metadata, or, for short,
meta-metadata.
Meta-metadata are data that describe (and in case of active
meta-metadata control) the operations of metadata. This is more
common than one might think. The definition of modeling languages
like E/R or UML are meta-metadata. Also, the modeling language
extensions of UML for data warehousing are meta-metadata.
You need to deal with meta-metadata if you want to integrate
metadata from different sources. This situation is the typical
tool integration scenario as found in all kinds of heterogeneous
tool environments.
For example, in data warehousing, you have to integrate metadata
from staging tools with metadata from the data warehouses and
data marts with metadata from the analysis systems.
4 Further Aspects
University of Zurich and SwissLife provide a technical report
that discusses further properties of metadata with a specific
focus on data warehousing. Please see: Martin Staudt, Anca Vaduva,
Thomas Vetterli. Metadata Management and Data Warehousing.
Report 21. Zurich, Switzerland: Swiss Life, July 1999.
|