Hi Martin, 2017-08-18 10:09 GMT+02:00 Martin Knoblauch <kn...@knobisoft.de>: > > > Lets say I wanted to increase MaxRequestWorkers to e.g. 800. One of the > several solutions would be to up ServerLimit to 32 and leave > ThreadsPerChild at 25. But I could also leave ServerLimit at 16 and up > ThreadsPerChild to 50. Or I could choose any combination where > ServerLimit*ThreadsPerChild is bigger than 800. What I would like to know > is which way to go. And why. What gives the best and most reliable results. > Assumption is that there is plenty of free memory, CPU and network > bandwidth available. >
my personal recommendation would be to prefer increasing the number of threads rather than the number of processes (threads are supposed to share more and be less heavy to context switch for the OS, but on Linux this can generate loooong discussions and flames so I will not go there :), but you'd probably need to perform some load tests with different configurations and see which one suits your system more. From the reliability point of view I don't see much difference between the two approaches, again testing would probably give you more datapoints to choose the best approach for you use cases. Interesting to read is also what mpm_event offers over mpm_worker and what use cases it solves: https://httpd.apache.org/docs/2.4/mod/event.html One thing that might also be good to check is the following: https://httpd.apache.org/docs/current/mod/mpm_common.html#listencoresbucketsratio When you read it please keep in mind that each process in mpm_worker/event allocates one thread as listener, so there is contention for the listen socket. Luca
