The Advancing Art of Indexing - High-Performance Web Databases: Design, Development, and Deployment - page 637

Client/server systems answered, at least temporarily, the next level of
end-user access issues. The point-and-click interfaces of Windows and
Macintosh systems made interacting with the data far easier. Users no long-
er had to memorize command sequences to get in and out of applications
and data bases. They could perform queries and create reports on the fly,
and download data directly into the PC. Advances in network technology
made it possible to have access to any number of corporate systems from
the same PC or workstation.
With client/server connections, middleware connectivity tools, and net-
working, IS has solved the second hurdle of providing universal access: us-
ers are online and can request information through a relatively intuitive
graphical environment. Fast access to information through online inquiry,
analysis, and reporting remains a crucial factor if today's knowledge work-
ers are to get their questions answered and work completed. But, as with
most elements of progress, this new access has brought with it another
dilemma — performance.
THE PERFORMANCE DILEMMA
Unfortunately, there appears to be one indisputable constant: fast is
never fast enough.
As users point-and-click their way to the data they need, both users and
IS now dread the infamous QFH (“query from hell”). Perfectly reasonable
business questions such as “How many customers in the northeast region
bought our product last quarter at the promotional price?” can bring even
the most powerful server with millions of records to its knees — and all
other users along with it. Because of this slow response time, interacting
with the data is clumsy at best and impossible at worst.
Without question, CPU-intensive tasks, such as engineering and scientif-
ic applications, have seen dramatic increases in speed. In addition, the
CPU-intensive tasks required in data base applications, such as sorting, are
considerably faster. With CPU speeds increasing and the cost per millions
of instructions per second (MIPS) dropping, it might appear that the solu-
tion to transaction and analysis bottlenecks experienced in most data base
applications has been met. Unfortunately, this is not true.
The reason is that most data base applications benefit only modestly
from higher-speed CPUs. Typically, the inability to resolve user queries
comes from the application being disk I/O-bound rather than CPU-bound.
The CPU is busy reading the data from disk in order to answer, or process,
the request. Two basic strategies have emerged in an attempt to solve the
performance dilemma:
• Limit the number, type, or timing of the queries that can be done
through the client/server system.