We have also witnessed less than perfect load balancing by
mod_jk using tomcat-3.2.3. We have four tomcats that we
are load-balancing across and the first one seems to always
get more sessions, with the second, third and fourth each
have somewhat fewer than the previous one. For example,
right now the first has 139 sessions, the second 73, third
66 and fourth 52.
My guess is that it has something to do with the round-robin
load balancing scheme and the multiple copies of apache httpd
that are running...perhaps each time a new httpd is forked it
starts the round-robin rotation over at server #1, and maybe
somehow new session requests tend to be handled by an httpd
that has just forked?
Anyway, when we used JServ in the same hardware configuration
we saw much more even load balancing of sessions. It makes
sense (to me) that a random server selection like mod_jserv
uses would work better when you have multiple unsynchronized
selectors than a round-robin scheme.
-Don
At 09:20 PM 9/22/01, "Cezar Grzelak" <[EMAIL PROTECTED]> wrote:
>I have a two tomcat boxes running tomcat-3.2.3 which are load balanced using
>mod_jk. Now everything worked fine until I started playing with the lbfactor
>parameter in my workers.properties. Initially I had it configured so that
>one box had an lbfactor of 3 while the other one had 2. When I tried to
>switch the lb factor to be 1 on each box my cluster no longer worked! Only
>one node on the system would receive all the trafic while the other would
>sit completely idle. Certain combinations of lb factors seem to work just
>fine (4 and 5 for example) while others seem not to work at all. I can't
>even determine the pattern for which lb factors work. Has anyone come across
>this issue before? It seem like a weird bug in mod_jk to me!
>Please help as I'm running out of ideas on what to try next. Nothing seems
>to cure the problem unless I set the lb factors to the values that mod_jk
>"likes".
