Why does the response time graph show large spikes?

These spikes can be caused by latency and / or routing anomalies anywhere along the route between the monitoring node and your server.

While our monitoring system utilizes multiple nodes in disparate locations to confirm detected failures (primarily to avoid false alarms) the graphs are produced from a single node, and hence subject to this variability.

One option we are considering to help avoid confusion is to offer a rolling hourly average of response times; which should provide a more useful picture of variations throughout the peak / offpeak cycle.