Squid now is serving as reverse proxy before apache cluster server
some php and jsp page must be cached in squid to increase the response speed and decrease the j2ee or apache thread number.
using refresh_patterns solved it.
refresh_pattern -i .jsp 720 25% 1440 override-expire ignore-reload
refresh_pattern -i .php 720 20% 1440 override-expire ignore-reload
ignore-reload will prevent request no-cache etc,like ctrl+F5 under ie will refresh the squid cache,but ignore-reload will prevent it.
ATT:
squid default squid.conf include
refresh_pattern -i .js 360 25% 1440, the above -i .jsp must stay top before this.else -i .js will be suit for the parsing.
Total web HA architecture under linux maybe like below
Load balancer: lvs(ha include),now LVS included in RHEL
Web Server: multi httpd server,now apache(file rsync or distributed file sytem or nfs)
Stream server: Helix or WMS
Other Service: smtp,ftp etc
Application server(j2ee Container): tomcat(javagroups,tangol),jboss,weblogic etc,now using mod_jk link web to app
Database:oracle rac,mysql(using clusterjdbc,no mysql cluster which now testing)
Cluster Filesytem: now testing lustre1.2.1 under sles9(Suse Enterprise server 9),but too less document
SAN too expensive and NAS not suite for so much server