Squid hiperformance

From NuclearCat's homepage

Jump to: navigation, search

[edit] Source code patches

  • Patch comm_epoll.c

When connections amount reach 2000 clients at moment, and you have large disk storage, you can face slow responce even for local resources (cachemgr). Possible there is not enough events buffer to serve them all. 

- #define MAX_EVENTS      128   /* max events to process in one go */
+ #define MAX_EVENTS      16384   /* max events to process in one go */

[edit] other

Recommended

  • 1000HZ
  • CFQ scheduler (in some cases deadline is better)
  • ulimit -n 10240 and higher
  • increase to net.ipv4.tcp_max_syn_backlog = 10240
Retrieved from "http://www.nuclearcat.com/mediawiki/index.php/Squid_hiperformance"
Personal tools