If parallel processing is a certainty for your fast-growing organization,
here are 10 criteria for judging whether an MPP-driven data warehouse is
in your future, suggested by the Standish Group.
Scalability. All aspects of the data warehouse must scale: the database
engine, storage, processors and the tools to manage the system. When scaling
becomes necessary, it must be accomplished in a smooth fashion invisible
to the end user.
Date Management Utilities. Database loading, data redistribution
and other routine management operations should take advantage of the parallel
environment, operating concurrently with online user queries.
Performance Optimization. To achieve high performance, time-intensive
CPU operations, such as sorting, must be minimized. The system also should
be designed to enhance the location of the data and to optimize queries.
Openness. In this context, open means able to support a range
of modeling and desktop analysis tools, through ODBC or other standard interfaces.
It also implies the ability to gather data from legacy systems and from
dispersed OLTP systems.
Availability and Integrity. Continuous availability (fault tolerance)
is the optimum solution. The system should also support user queries of
varying length and complexity in addition to data management utilities.
Data integrity, meanwhile, is a prevention issue. Again, fault tolerance
is the best solution. Minimally, that means RAID storage technology and/or
Mixed Workload Support. The system should shield users from performance
variations associated with the processing of queries of varying length and/or
complexity and provide the appearance of near immediacy for all users.
Cost of Ownership. Be sure to evaluate not only initial price/performance
metrics but also the cost of expansion--additional processors, disk storage,
operating software, utilities and support.
Integration. Vendor expertise can enhance the opportunity for successful
installation and minimize the risk of failure. First-timers should consider
vendor-supplied consulting to be essential.
Technical Experience. Look for and verify customer references with
similarly sized data warehouses from the proposed provider.
Industry Experience. The ideal vendor will have the industry-specific
experience to guide first-time users to the "low-hanging fruit"
that yields early, impressive benefits.