In a “flash crash” world, is your site ready?

(Editor’s note: Imad Mouline, CTO of Gomez, the web performance division of Compuware. He submitted this story to VentureBeat.)

During May’s stock market “flash crash,” websites of some of the nation’s largest brokerage firms slowed to a crawl, taking nearly a minute to load – something that’s unacceptable in a world where customers get impatient after only four seconds.

If these major websites can buckle, how can a startup maintain fast website performance under unusually high traffic?

New web properties are especially vulnerable to this problem because they haven’t built a loyal following, so visitors are more likely to go elsewhere if the site is performing slowly. In a recent survey, 78 percent of consumers said poor Web performance at peak times has driven them away.

While the “flash crash” is an extreme example, any website can experience a peak load event. The causes can be external, such as a high-volume news day (Michael Jackson’s death) or shopping day (Cyber Monday), or a regional ISP slowdown affecting big chunks of your audience.

More often, though, the causes are internal – and the result of poor planning.

Planning for peak loads is a challenge for startups because it’s difficult to predict traffic when your site has no track record. While the cloud may seem appealing and convenient, there’s a downside – placing your site in a shared environment relinquishes a certain measure of control. It’s important to understand how virtualized and cloud-based infrastructure impacts application performance, especially during peak traffic times.

To avoid a Web traffic overload, here are a few common (but key) problem areas – and potential solutions:

Regional performance – There can be significant differences in site performance across different geographies, with response times generally better based on proximity to the server. So peak load hours may be manageable in New York, but catastrophic in Seattle.

Solution: Monitor on a regional basis and compensate for the slowest regions. Once you’re comfortable with regional monitoring, take it a step further and monitor on a city-by-city basis. You may find persistent slow spots for which you can optimize or perhaps engage one or more content delivery networks.

Third-party content – Since many sites are mashups, your site can be hurt by the poor performance of third-party content. Even original content sites usually have several outside elements such as ad servers, video feeds or diagnostic code.

Solution: Monitor the impact of each one and demand performance service level agreements from each provider. It should be noted that some third-party content providers are cloud-based, which as a shared environment could make your site more susceptible to performance variability or Internet congestion.

Browser performance variation – When Internet Explorer was the dominant web browser, optimizing applications for that one browser could suffice. Now there are four or five major browsers, each with multiple versions, and many niche browsers.

Today’s modern browsers also increase the number of parallel host connections to achieve faster load times. This can double or triple the server load of visited sites and actually result in a site slowdown.

Solution: As cumbersome as this may be, you have to optimize your site for all major browsers and perform testing based on this increase in server load. Of course, first optimize for browsers that generate the most traffic or revenue.

Complex applications – Today’s web audiences demand video, interactive features and engaging web content. Many of these features now depend on sophisticated programs that run on the client side – within the browser of your customers. During peak traffic, these apps can slow considerably or break down.

Solution: Pre-test and monitor these applications’ performance as users see them. Testing from within your firewall may incorrectly show that all is performing well.

The bottom line is clear: Simple load testing from within your firewall or from a handful of data centers isn’t enough. These traditional methods might tell you how your infrastructure will fare under a simulated load, but they will never give you insight into how your end-users will experience your site under a similar load. To avoid peak traffic slowdowns it is vital to take an end-user perspective on website performance load testing.

Companies with consistent performance – the ones who handle peak load traffic effortlessly – monitor performance 24/7 from their end-users’ point of view. They sidestep the potential roadblocks occurring between the data center and a customer’s browser. They also continually optimize to compensate for the ever-fluctuating tendencies of their audience and the inconsistencies of the Internet itself.

That’s why, among the brokerage sites that slowed during the “flash crash,” there were a select few that maintained their usual performance levels.

You don’t have to be held hostage by external events. With proper internal planning you will not only avoid peak load meltdowns, but your new web property can enjoy a competitive advantage by performing well when other sites in your category are not.

Next Story: Crisp Wireless demoes cross-platform ads with first campaign
Previous Story: Kleiner, Sherpalo add $8M to mobile ad network InMobi

Tags: cloud computing, website performance

People: Jerry Buhlmann