Good Practice Guide for Developers of Cultural Heritage Web Services
Good Practice Guide for Developers of Cultural Heritage Web ServicesAuthor: Peter Dowdell, UKOLN
Originally commissioned by the NOF-digi
Technical Advisory Service for projects participating in the NOF-digitise programme.
This section provides advice and guidance on performance indicators for Web sites. The information provided in this section is intended to be initially used during the planning stages for a Web site and throughout the Web site's life.
The basis for generating the performance indicators will be the Web server (HTTP) log. Projects may be required to have the ability either to create reports directly from their Web server logs or to have them generated for them by competent Web hosting companies. A project's programme manager will ask projects to extract and forward a reduced set of data as part of the ongoing monitoring process.
The four measurements that a project's programme manager may require them to supply are as follows:
Explanation of these measurements are provided below:
| name | platform | calculates session length |
licence |
| Webtrends | MSoft | YES | commercial product |
| Analog | ALL | NO | Open Source |
| Awstats | Unix | YES | Open Source |
| Webalizer | ALL | NO | Open Source |
The number of distinct user sessions chalked up over the reporting period.
User sessions are determined by grouping together all requests that come from the same IP address within a time interval of no less than 30 minutes between each request. A figure of 30 minutes is widely used and is the default in many Web analysis packages.
Note: This is NOT the "unique users" value. User sessions include repeat visitors to your pages.
Note: Due to issues with NAT (Network Address Translation, used when several machines are connected from a local network through one Internet connection), proxy servers, etc there may be times when many users are simultaneously using the site all through the same IP address. In this situation, the user session becomes invalid - all these user sessions will be counted as one huge user session. This is a limitation of the way the Web works, and all services which collect statistics are subject to the same distortion of their figures.
By examining the time of the first and the last request made during a user session, a figure for the length of the user session can be obtained. The average duration of a user session is the average length of all user sessions found.
Note: some good packages (notably Analog amongst others) are not able to calculate the user session length, so cannot provide this information. Where this is the case, projects should answer with "n/a : " followed by details of the log analysis package in use.
Total number of requests for files that are defined as pages.
Generally files that have extensions .htm, .html, .shtml, .php, .asp, .pl, .cgi and so forth. The exact set may differ amongst projects, projects will be expected to set up their analysis packages so that all page-type files are measured.
Note: Do not include images, graphics, stylesheets, external script files or other "component" files that together comprise one page.
Note: Requests that can be positively identified as emanating from non-human sources should as a matter of standard practice be excluded from the analysis. You must ensure that your reports do not include data coming from search engine spiders, network monitors, benchmark tests and other generally invalid sources.
Note: Some projects may be generating dynamic URLs where the
base URL remains the same, and the page delivered is determined by the
contents of the query string: i.e.
http://www.domain.com?page=search&type=quick.
In this case, special steps may have to be taken to convert
log files to sensible page impressions values.
The average is obtained by dividing total page impressions by the total number of user sessions recorded.
The following background information may be useful.