home | software | site monitor | vizzavi case study | charts
Plotting the logged data can highlight interesting and difficult to detect trends. These charts are quite self explanatory. I've started a simple tool to pre-process the logged data, which will make these charts somewhat less cluttered.
These charts represent data collected over approximately five days in April 2002, with the application running on my home computer in London communicating via ADSL. More meaningful results would be obtained by collecting data over a longer period, and by using the pre-processing I mentioned above. It would be worth collecting data for more than three user actions (login, check mail, logout) as well.
![[chart]](VizTestGraph02.gif)
Login
Check Mail
Logout
This chart shows responses that returned in zero to thirty seconds over the indicated time period. You can notice repeating patterns. Checking of mail is slowed at 6PM and 8PM daily, no doubt due peaks in usage. Interestingly, logout (yellow) peaks between 6:45-7:56 AM, perhaps due to maintenance. Following graphs highlight these patterns more effectively. It also appears that login (blue) is increasing steadily.
![[chart]](VizTestGraph03.gif)
Login
Check Mail
Logout
This chart is identical to the previous one, though it shows response times up to 100 seconds. We see Login (blue) peaking at the same time as the above-noticed logout peak (between 100-150 sec) occuring between 6:45-7:56 AM every day.
![[chart]](VizTestGraph04.gif)
Login
Check Mail
Logout
In the above chart, each day's results are plotted over each other. This approach highlights daily trends more effectively that in the charts above. The solid black line is a moving average line, using a period of 60 seconds. On this chart, we clearly see the early morning logout and login delay, as well as the evening peaks in mail checking. The other average lines sit between 0-10 seconds, and are exploded in the next chart.
![[chart]](VizTestGraph05.gif)
Login
Check Mail
Logout
This chart is quite a mess -- it would really gain from the data pre-processing mentioned earlier. There appears to be a small peak in checking mail (pink) between 1-2 AM, however the times are still very acceptable.