When they do,
they immediatley discover that some form of optimization is required.
Information Optimization has to be built in, each step of the way. It
usually cannot just be tacked on at the end of the process or fixed by
buying a bigger server.
- Focus - Build each system
to accomplish a discreet set of business requirements. This keeps components
fast by letting each do the tasks that it was intended to do.
- Use De-normalization to
reduce joins whenever possible.
- Use maximum indexing to
increase query speeds whenever possible.
- Pick the right vehicle for
each purpose. Do not try to build time series analysis with SQL and
a report writer. Use MOLAP tools that are engineered to do this work
for you.
- Minimize the less important
data elements. An 80/20 approach. Much of the data that feeds your system
will never be analyzed. Choose what to include based on business priorities
and then consolidate the less important categories as "Others"
where you can.
Samples:
- Create focused retention
instead of retaining everything. Consider data retention.
- Use a separate reporting
instance where OLTP systems have no time left for report queries.
- Individual Datamarts for
each logical business unit.
- Create separate application
type for each user type. (OLAP, Ad-Hoc, Guided, Canned, etc.)
- De-Duplicate early in the
process.
- Use database loaders to
get a flat file quickly into a DBMS. Then use DBMS performance to do
the work.
|
