Correction: response time (%D
) is μs not ms! 1
This doesn't change anything about the weirdness of this pattern but it means that it is practically way less devastating.
Why is response time inversely correlated to request frequency?
Shouldn't the server respond faster when it is less busy handling requests?
Any suggestion how to make Apache "take advantage" of less load?
This pattern is periodic. That means it will show up if impressions drop below about 200 requests per minute - which happens (due to natural user-activity) from late night to early morning.
The requests are very simple POSTs sending a JSON of less than 1000 characters - this JSON is stored (appended to a text file) - that's it. The reply is just "-".
The data shown in the graphs was logged with Apache itself:
LogFormat "%{%Y-%m-%d+%H:%M:%S}t %k %D %I %O" performance
CustomLog "/var/log/apache2/performance.log" performance