Home > Reference > Articles

Understanding web performance graphs at a glance

Most web managers may be familiar with the types of performance graphs shown on this page, which give a snapshot of how the site is performing. However, what exactly do they tell us and how can they warn of problems ahead?
Paradoxically, the more successful a website, the greater the risk of reduced performance at peak times as bottlenecks develop. Sometimes the success of a site can be so dramatic that it catches everyone by surprise. Regular testing is perhaps the most effective way to help keep pace with any increase in visitor volume.

Performance graphs illustrate the site's effectiveness at a glance by showing when a page or transaction is taking an unreasonably long time to respond. The graph's shape can indicate possible bottlenecks now and give an indication of problems ahead.

It is probably realistic to assume that most site users are happy to wait a few seconds for their clicks to bring results, but conversely after a prolonged period, say 30 seconds, they will become impatient or may lose confidence in the site altogether. Hence, a performance graph that has lots of peaks above say 10 seconds should cause concern and if there are a lot of readings off-scale and well above the average - 'outriders' - these should definitely ring alarm bells. It probably means that your site can no longer cope with the number of visitors it attracts and that most visitors are finding it a less than satisfactory experience.

Each performance graph plots the time to carry out a transaction or download an html page (in seconds) against the time of day, and will normally show the pattern over a few days. The response times, in green, are supplemented by a red bar whenever there is an HTTP error ie the user could not access the page or transaction at that time. The lower red peaks below one second, represent the time to connect to and receive a response from the website's server before the HTML download begins.

Look out for these four patterns

At SciVisum, we've identified four broadly different patterns that occur regularly which tell us at a glance how well a page or transaction is coping with traffic and whether there have been times when it has been inaccessible. Below, to keep the charts manageable, we have chosen a Y-axis cut-off point around 10 seconds. The resultant chart shows the general pattern, but cuts off any major outriders. However, these do influence the smoothed average line, plotted in blue, which takes every reading into account. Any 'bumps' in the blue line indicate a high degree of large outriders.
The examples below are fairly representative of the patterns we look for. We've likened them to different wave states - bringing a whole new connotation to surfing the web.

Choppy

The worst performing graphs display a choppy appearance. The majority of readings are outriders. HTTP errors account for between 1% and 10% of all timings.

Even the smoothed average will exhibit obvious peaks and troughs. Sites falling into this category will continually struggle to meet demand. The end-user will experience inconsistent performance, slow response times and many errors. The site's ability to retain visitors will be seriously impaired and, despite far more visitors, sales may drop substantially as the wheels come off the shopping trolley.
When a site reaches this stage, SciVisum recommend functional stress testing coupled with an engineered analysis at several levels including network, HTTP, web server configuration and application functionality.

Wavy

Slightly less erratic, more values grouped around the average, but still a lot of outriders and some HTTP errors. Outriders and HTTP errors occur about 1% of the time - this may sound acceptable, but it represents about 15 minutes when visitors are unable to add to their cart every day, and most errors will occur during the peak period for sales.
The smoothed average exhibits a pattern of small peaks and troughs - the occasional swell, which tells us that the site may work well a lot of the time but will regularly encounter bottlenecks, maybe at exactly the wrong times.

Gentle Rollers

The majority of readings are grouped around the average download time. You can still see a number of outriders and maybe some HTTP errors. However these are generally less frequent, perhaps 0.1% to 0.5% of the time (1.5 to 7.5 minutes a day). Far better, but perhaps signifying that the system is beginning to encounter bottlenecks, which will degrade performance as usage increases.

The lightly rolling waves of the smoothed average line are acceptable, but may imply that the site is becoming less predictable. To many users inconsistency can be as bad as being constantly slow. Purchasers, uncertain about what to do next, may think that the site has crashed. However ,in our experience the problems can usually be overcome without the need for new hardware or acceleration software by identifying the causes of outriders and then taking steps to optimise the system.

Calm

A fast, consistent site normally throws up a performance graph with few outriders and HTTP errors. The graph is very smooth, the smoothed average virtually a flat, straight line. This is the ideal, indicating that the system handles the load well and responds to demand consistently. All such sites need is to monitor functionality 24/7 to identify future weaknesses at an early stage.

In a Study carried out over a four-week period in 2004. SciVisum monitored the "Add to Basket" function across 51 e-Commerce sites using our SV-Monitor 24/7 monitoring and functionality testing software.

80% gave an inconsistent performance with varying response times and HTTP errors. In 53% of cases we found a choppy or wavy pattern. These sites regularly encountered bottlenecks, varying response times, timeouts and errors.

27% of performance charts were classified as 'Gentle Rollers' - their sites provided a reasonably consistent experience, but with errors and unpredictable response times on occasions.

Only 20% had a smooth performance graph which might suggest that a minority of sites provide shoppers with the best possible experience.
By adopting SciVisum's surfing analogy to performance graphs, web managers can assess their performance at a glance and by correcting any slowness might just avoid a total wipeout at peak times.