Over the past three years as an Allaire partner, I have been involved in countless marketing presentations where I touted ColdFusion as the best solution for producing Web-based applications. During this time other product vendors, instead of selling against CF based on feature comparisons, have instead focused on peripheral issues. Throughout the CF 2.0 product cycle, I would routinely be questioned about Allaire's staying power in the market, financial performance and capital inflows. Allaire's rapidly expanding market share and IPO silenced those critics. During the CF 3.x product cycle, our competitors shifted to attack CF as an unsuitable product for the enterprise, questioning whether it could "scale" to support an unlimited number of concurrent users.

Despite anecdotal evidence to the contrary (AutoByTel and SmartMoney interactive), some clients remained unconvinced. Allaire recognized this tactic and proceeded to load CF 4 with features to address this concern. Chief among these is the Enterprise edition's BrightTiger ClusterCats integration. Certainly ClusterCats has been an invaluable marketing tool for positioning CF as ready for the enterprise. However, one must ask whether using clustering eliminates the need for designing software that functions efficiently. Throughout my experience as a scalability consultant, I have yet to encounter a Web site that could not be made to scale by tweaking the underlying source code. Optimizing CF applications should be viewed as a five-step process.

Step 1: Review the Debug Stream
Reviewing the debug stream continues to be the fastest method for identifying slowly executing pages. A total page execution time over five seconds may indicate a problem. Review the queries that are executed on each page. Are you retrieving the same data each time? Try caching the query for enhanced performance.

Step 2: CFML Code Review
New CF developers, unfamiliar with the vast number of functions, choose to code their own, less efficient methods in CFML. Granted, reading the CF function reference may not be as exciting as a Grisham novel, but it's necessary. Using the valuelist() function to generate a comma delimited list from the values in a given query column instead of coding a concatenation algorithm using CFLOOP can shave seconds off your execution time. A list of other CFML coding tips can be found on the Allaire Web site at http://allaire.com/Handlers/index.cfm?ID=1576&Method=Full.

Step 3: Leverage the Database
Poor database design runs rampant in our industry. Review your data model using a database design tool such as ER/Win from Platinum Software. Are you enforcing referential integrity? Is the database represented in at least a third normal form? Consider the queries you'll need to execute throughout your application and ask yourself how easy (or difficult) it will be to retrieve the information from your data structure. Next, move all your queries into SQL Stored Procedures. CF 4.01 now supports returning record sets from ORACLE and passing parameters to MS Access Queries through the <CFSTOREDPROC> tag. You'll get your data returned much faster than using traditional SQL passthrough.

By using Stored Procedure program constructs you may also be able to combine several sql statements into a single call, eliminating the overhead associated with multiple queries. Still seeing poor response times? Review your database's SQL reference. Query execution time may be reduced by up to 90% by utilizing aggregate functions and platform-specific SQL constructs that quickly perform calculations on vast sets of data.

Step 4: Leverage the Client
Many developers irrationally fear using client-JavaScript in their applications. While support for the language does vary widely among the browsers, a high payoff can result from transferring application logic from the server to the browser. End users see better response times since no server request is involved and the traffic between the browser and server can be greatly reduced. These high-payoff items outweigh the slightly longer development and debugging cycle that JS development requires.

Step 5: Review the Hardware Configuration
Once you've optimized the application programmatically, review the CF server's hardware and software configuration. Make sure that the application and database servers are on different machines and have adequate amounts of RAM (512 MB and 1 GB, respectively) if you expect a lot of traffic. Allocate enough memory for the CF template cache to allow the server to cache all of your application pages. If you've gone through this entire process and are still experiencing server timeouts or require failover capability, then it may be time to take the clustering plunge.

The continual advances in processing performance have allowed developers on other platforms to write bloated, inefficient code. I view ClusterCats, mistakenly seen as a panacea for scalability, as potentially having the same effect on CF application development. Stop the madness before it begins.