Next Page | Prev Page | Main Index | Home |
The business is an online reference service. The problem is to
be able to track the cost and income streams associated with billions
of documents across millions of users. The source of this information
is about 15 million transaction records per day. In addition to
user and document numbers, many different subtotals are needed
by such things as, sales rep, state, database, library, market,
tier, file, branch, country, firm, affiliate and service.
The current solution is based on a whole lot of batch processing.
The first step is to filter and summarize the raw input on a nightly
basis (information loss #1). The next step is to produce/update
a set of databases IDMS, DB2, and others (information loss #2).
Once these databases are ready a subset of information can be
extracted from each (information loss #3). These extracts are
processed monthly to produce a multidimensional database (information
loss #4). Finally, questions can be asked. But only of information
that actually makes it into this multidimensional database and
only on the monthly snapshot.
In contrast, the MPbase solution contains four parts:
1: Load the raw data directly into an MPbase. This is possible
because the nature of MPbase allows it to accept high volumes
of relatively unfiltered data. The process of loading tends to
highlight any errors in the information. This allows for the isolation
of errors without the normal information loss.
2: Set up MPbase views to feed and/or support the IDMS,
DB2 and other database systems. MPbase can appear to be
almost any type of database. This allows two options in dealing
with the currently running systems. The first is to produce an
extract in the optimum format and order to load into the current
database systems. The second is to create an MPbase view
that allows the application to think it is still talking to the
old database.
3: Load any plans and forecasts into the same MPbase. MPbase
has the unique ability to deal with different levels of information
in the same database at the same time. This allows direct comparison
of raw detail with planned totals. It is only necessary to load
the lowest level planning numbers as MPbase can produce
the higher level totals internally. Currently where there are
separate weekly and monthly numbers, daily numbers, if available,
could provide both.
4: Use the MPbase directly to answer the same multidimensional
questions in much more detail. Questionable records may be included
or excluded in the viewed totals. Questions would be answered
with real time data with no delays waiting for batch work to complete.
Any total at any time could be drilled down to its lowest level
components. This puts an end to the long and tedious process normally
required to "prove" a number produced from a traditional
multidimensional database system.
In short, faster access to better information with MPbase
this nasty real world problem is no problem.
Next Page | Prev Page | Main Index | Home | © 1998-2004 NPSI |