Re: ...Java components

I'm not an expert here, but I have dinked about with a couple of commercial server-side java systems (Jrun and Dynamo). They both have a standalone server process with a small glue module stuck into Apache to pass requests. This is analogous to database connections. It appears to me that this approach miminizes the problems discussed here, in particular the memory footprint and JVM startup overhead. However, the java engine consistently used up 5-12% of CPU just idling, sigh. G Begin forwarded message: [snip] In early design discussions with a developer at Sun, he recommended against running the JVM inside the Apache process under Unix. Issues with threading, I/O, exceptions, set/longjmp, etc. were cited I was preparing to add logic to optionally fork a process, but this data makes it look like somebody within IBM already came to the same conclusion and implemented the logic behind the interface. Note: your data suggests that each httpd process will eventually become accompanied by three slave processes. (Plus you have one extra: presumably this is the grep itself). These additional processes should not change the number of times you saw the initial startup costs associated with launching the JVM. P.S. on my configuration, instead of the apache processes round robining the requests, the same one gets to serve all requests, presumably until such time as a second, concurrent, request were to ever show up.