Pattern 42:
AKA:

You know that performance, user impatience, their lack of free time, work pressure and their perceived cost of communications will affect their attitude to your site. Since you have ESTABLISHed THE USE CASES (3) you can deduce the kind of download experiences they are likely to have. Since you have adopted AUTOMATED TESTING (6) you want to test and collect metrics on download and display times. Everyone wants a sticky site.

Users leave if download times are not short, especially on dial-up lines. Users who leave a site in dismay will usually never return. For any site this is undesirable, for commercial sites it is catastrophic.

Therefore

Keep the page sizes small enough that download times are less than 10 seconds on a 28K modem using SHORT TEXTS (44). Avoid nested tables that will slow downloads. Provide FEEDBACK (41) and a SENSE OF PROGRESS (48) using progress bars and pop-ups prior to download, using THE RHETORIC OF ARRIVAL AND DEPARTURE (20). Make sure both the physical and subjective download time is tested, including on TWO YEAR OLD BROWSERS (10). Use AUTOMATED TESTING (6) too.

To help reduce subjective time using THE RHETORIC OF ARRIVAL AND DEPARTURE (20). DESIGN PAGES FOR SCANNING (43) and use only SHORT TEXTS (44). Print CONTENT BEFORE GRAPHICS (55).

Contributors and sources
Mattias Larsson.Veen (2001) contains a interesting chapter on speed.

The size of a file in kilobytes (don’t even think about megabytes) gives some indication of the download time it will need. However, this is only approximate and is unpredictable due to variable network latency, site traffic and error correction. One of the worst sites in this respect that we have found is at www.thetimes.co.uk, which varies from almost instantaneous download to several minutes – apparently dependent on the time of day. One deduces that testing for this is not done, rather than that the designers are so underpaid and abused that they just don’t care.

One can carry out usability testing: going through the use cases with a stop watch or collecting timing data using a product such as Speed-trap’s Prophet.

Perceived time is not always the same as actual time. If the user has something interesting to do or look at during a down-load she may be more tolerant. The overhead of download an animation may pay for itself in this way, but it must be useful and relevant to the task.

To reduce latency caused by the need to establish a new connexion for each object make sure that the server supports HTML keep-alive.

Browse the language	What is Wu?	Look at an example pattern sequence	Structure of the patterns
Comment on Wu	Contributors	Return to TriReme home page	Links to related sites