By definition of caching you're not going to see a sudden magical increase in speed on your site. You need to enable it, and wait for awhile so that the caches can get built up by people viewing the site and start actually being used as the same things are viewed over. You may also want to try an optcode cache like eAccelerator, APC, or xcache if you don't already have one.
However, by your mention of the MessageCache and $wgUseDatabaseMessages, the issue looks like it has to do something with the database queries used to grab the messages.
~Daniel Friesen(Dantman) of: -The Nadir-Point Group (http://nadir-point.com) --It's Wiki-Tools subgroup (http://wiki-tools.com) --Games-G.P.S. (http://ggps.org) -And Wikia ACG on Wikia.com (http://wikia.com/wiki/Wikia_ACG)
Neil Bird wrote:
In a nutshell, I've upgraded a previously working 1.5.(1?) install to 1.12.0. It seemed to go smoothly, and everything's there with no discernible errors, but pages are taking some 20 s. to come back.
I'm not running any caching, but there are 4 wikis on this [CentOS 5] box running under Apache virtual servers, one at least of which has been 1.12.0 for a bit and doesn't show any slowness with no caching.
I've compared the LocalSettings.php files for the OK one and my slow one, and there's nothing obviously different that might affect this.
Bizarrely, we have a development box alongside the production one which ought to be identical (they actually both virtual boxes on the same hopst(s) just to confuse matters), and my wiki runs at a decent rate on that one. Same versions, I've copied the d/b across (mysqldump --add-drop-tables), and compared the entire 'wiki' subdirs. No obvious diff.
The only advertised diff. between the production box and the dev. one is that the production box uses SELinux.
*Also*, I have found that I can ‘remove’ the speed problem by using “$wgUseDatabaseMessages = false;“ (from a google around the issue). However, the MediaWiki docs imply that this would be obviated by using caching, and I have tried memcached temporarily, and it seemed to make no difference at all.
I'm now down to analysing a strace of a php session between the working wiki and mine.
Just for completeness, here's a bit of my strace that displays a lag. Note that I can find no ref. to ‘Revision.php’ in the strace of the working one. Also, just prio to this there's mention of ‘MediaWikiBagOStuf...’ and then ‘MessageCache::loa...’ in what I think is a channel to mysql. There's no mention of MessageCache, bar one to the .php file, in the working strace.
Does anyone have any ideas about what I can look at next?
time(NULL) = 1214560045 lstat64("/var", {st_mode=S_IFDIR|0755, st_size=4096, ...}) = 0 lstat64("/var/www", {st_mode=S_IFDIR|0755, st_size=4096, ...}) = 0 lstat64("/var/www/fwwiki", {st_mode=S_IFDIR|0770, st_size=4096, ...}) = 0 lstat64("/var/www/fwwiki/wiki", {st_mode=S_IFDIR|0775, st_size=4096, ...}) = 0 lstat64("/var/www/fwwiki/wiki/includes", {st_mode=S_IFDIR|0775, st_size=12288, ...}) = 0 lstat64("/var/www/fwwiki/wiki/includes/Revision.php", {st_mode=S_IFREG|0664, st_size=22430, ...}) = 0 open("/var/www/fwwiki/wiki/includes/Revision.php", O_RDONLY) = 4 fstat64(4, {st_mode=S_IFREG|0664, st_size=22430, ...}) = 0 lseek(4, 0, SEEK_CUR) = 0 read(4, "<?php\n/**\n * @todo document\n */\n"..., 8192) = 8192 brk(0x8c54000) = 0x8c54000 read(4, "er = isset( $row['user'] "..., 8192) = 8192 read(4, "o external storage if required\n\t"..., 8192) = 6046 read(4, "", 8192) = 0 read(4, "", 8192) = 0 close(4) = 0 nanosleep({1, 213780000}, NULL) = 0
<snip lots of these> nanosleep({1, 70533000}, NULL) = 0 access("/var/www/fwwiki/wiki/serialized/MessagesEn.ser", F_OK) = -1 ENOENT (No such file or directory) access("/var/www/fwwiki/wiki/languages/messages/MessagesEn.php", F_OK) = 0 stat64("/var/www/fwwiki/wiki/languages/messages/MessagesEn.php", {st_mode=S_IFREG|0664, st_size=188674, ...}) = 0 time(NULL) = 1214560066