After looking at a number of database performance-tuning software packages, including Oracle Enterprise Manager and Quest Software's Spotlight on Oracle, Treadway chose Confio Software's Ignite for Oracle.
Using the tool and its "wait-time analysis" features, Treadway was able to quickly pinpoint three pieces of SQL code that were causing half of the database's delays. For example, Treadway was able to get one piece of code fingered by Ignite rewritten so that a process that formerly took four minutes would run in just 10 seconds.
"If it runs 15 times an hour all day long, you can see how that can accumulate," he said.
Since starting to use Ignite a year ago, Treadway has been able to improve database performance by 30%. He says he could have wrung a 50% boost out of the system with further tweaks, but he didn't bother because the company is about to move from 8i -- an eight-year-old product for which support has ended -- to Oracle 10g R2.
But even so, Treadway figures that by using Ignite he'll be able to better test the new Oracle database's performance as he rolls it out, rather than "just make sure it works, which is what I would have done without Ignite."
Building a database-backed Web site the right way
Backing a dynamic content or e-commerce Web site has become one of the most popular applications for databases in recent years. Yet, the vast majority, at least according to Neil Day, former CIO at Walmart.com, aren't built in an efficient way.
According to Day, in the typical three-tiered Web application architecture, you "beat the crap out of database." To prevent a bottleneck at the database, DBAs need to "insulate the database" from the application's requests by using a combination of caching and very intelligent balancing of traffic.
To do that, Walmart.com created nodes that were each made up of four lightweight, commodity PC servers.
"There wasn't a piece of Cisco, Sun or EMC gear -- it was all cheap x86 boxes and other disposable hardware," he said.
With the entire application stack including the massive Walmart.com database, each node is able to cache and handle all of the traffic requests a particular user might make. That minimizes bottlenecks and the resulting delays, and it ensures that a user's requests always stays local.
According to Day, Google's strategy of replicating its search index database over tens of thousands of PC servers for speed and redundancy, called sharding is a similar, less complicated version of his strategy.
But Day says that sharding doesn't work as well for media content sites or an e-commerce site like Walmart.com, which has a huge inventory of items as well as millions of graphical elements and images, and as a result has a much larger database than Google has.
"The interesting part is how to make sure that most of the data any particular user will need is always near those four servers," he said.
Day is loath to reveal many more details. But using this strategy, Walmart.com was able to become one of the largest e-commerce sites in the world while "running on a couple of million dollars worth of hardware," he said.