Hmm. That's not even clearly indicative of an MHD-related issue. What you should look at is the buffer size MHD offers when calling recv() on the TCP stream. The size of the buffer MHD offers to the application can be smaller simply because the OS didn't receive 13k data chunks between calls anymore -- be it because the sender is slower, TCP had a congestion event or your receiver happens to be draining the buffer slightly faster. So there are many _other_ reasons that could explain a decrease in upload_buffer_size.
So to diagnose this better, you first should check if the cause is really the size of the IO buffer offered by MHD to the kernel. Given that this is all for the same connection, I doubt this is the cause --- unless your application again doesn't always handle 100% of the data in each call, then you could of course also see such patterns emerge. Happy hacking! Christian On 11/04/2015 10:49 PM, Olivier Delhomme wrote: > I have made two graphs: one that shows the number of time the callback > is called (column 2): > http://src.delhomme.org/download/libmicrohttpd/number_of_calls.png > > And one that shows the mean value of *upload_size (column 3): > http://src.delhomme.org/download/libmicrohttpd/mean_upload_size.png
