Its constantly writing to the HD isnt it? and im sure mp_logmessages = 1 doesnt help either.
-PM
Sindre wrote:
First of all, harddrive performance doesn't really affect hlds-performance, except when changing maps. So just make sure DMA is enabled, and you're set to go. Second, what's this "1000 Hz" everyones talking about? It just doesn't make any sense without specifying.
- Sindre
visit:===== Original Message From Matt Heler <[EMAIL PROTECTED]> ===== -----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Ok I've seen on this mailing list some tweaks by people.. Some good , some bad.. some just plain crazy and dumb..
First off I'll give a few basic ones then get to advanced stuff later on ( probally in another e-mail to this list ) ..
- - -- Hard Drive --
First and foremost let me start with the slowest portion of a server, the hard-drive. Basically in a nutshell hdparm is your best friend for this scenario, use it if you have an ide drive ( if you have scsi then don't worry about this since scsi is vastly superior to ide). To check your ide settings run hdparm like this hdparm /dev/hda, this will bring up the current settings of your ide drive. (mine look like this )
tribe BNR2 # hdparm /dev/hda /dev/hda: multcount = 16 (on) IO_support = 1 (32-bit) unmaskirq = 1 (on) using_dma = 1 (on) keepsettings = 0 (off) readonly = 0 (off) readahead = 16 (on) geometry = 158816/16/63, sectors = 160086528, start = 0
To benchmark your drive before you apply tweaks to it , run hdparm -tT /dev/ hdX for me I get the following when I run that Timing buffer-cache reads: 1164 MB in 2.00 seconds = 581.03 MB/sec Timing buffered disk reads: 274 MB in 3.01 seconds = 91.13 MB/sec
Now for note, you optimally want your multicount and readahead to be the same ( hdparm -a and -m control theese seettings ) ( the higher the # for theese the better , but each drive has its own multi-count and readahead limit so use hdparm -iI /dev/hdX to determine what it is ) , and for increasing io and preformance you want unmarkirq and dma ON.. To adjust theese on your ide drive simple issue the following for dma
hdparm -a1 /dev/hdX
and then for unmarkirq
hdparm -u1 /dev/hdX
for note you can combine all theese flags into one line , like this
hdparm -a1 -u1 /dev/hdX
(fairly e-z so far ) .. Now most of thoose settings that I gave you are the safe settings , they generally work on most modern ide drives. Now if you want to tweak more out of your drive you can issue the following ..
for write back caching issue ( this will increase the performance of writes to the drive )
hdparm -W1 /dev/hdX
another is turning off , or setting acoustic management to its fastest setting ( most newer drives now have this feature , if your running a desktop leaving it on the default setting will make the drive quiet , but for a server its useless and turning it to fastest will increase the io and allow the drive to operate at a faster rpm speed )
hdparm -M254 /dev/hdX
Now run hdparm -tT /dev/hdX and check to see if your drive is faster ( it should be much faster, then it was previously ). You can go further then this with hdparm and play with more settings on it, but any other settings then thoose I mentioned can gernerally be fairly dangerous and wreck havoc on your drive and the contents within it.. O yes I forgot to mention that after you reboot your server, you will need to re-apply your tweaks , since on reboot the ide drive will revert back to it's normal default settings.
- - -- FileSystem --
Ok the next few tweaks basically vary depending on the filesystem your using. If your using resierfs ( like I'm ) ... it's fairly known that if you want the most performance you must mount your filesystem with the notail option. However if your using ext3 ( like most redhat users are ) and have a new nice and shiny 2.5 or 2.6 kernel then its also a known rule to add htree support to your ext3 filesytem to speed things up.. I'll go into detail on how to add this stuff .. but frist let me go over reiserfs ( since I use that , and I like it ).
In a nutshell , your going to want to open up your /etc/fstab and modify your your mount settings for reiserfs .. My recommended settings are to turn off tail support and to also turn off access time updating. To turn theese off change the follow example line like this.
if your reiserfs partiton looks like this.. /dev/hdb1 /tmp reiserfs defaults 0 0
your going to change it , to look like this /dev/hdb1 /tmp reiserfs defaults,noatime,nodiratime,notail 0 0
With theese off, you should notice file deleting , creating and accessing will seem speed up and or on roids =P
As for ext3 , there are several different tweaks and a few of them depend on what kernel your using.. If your using a newer 2.5 or 2.6 kernel ( or a modified 2.4 kernel with htree support ) and have the latest e2fsprogs packaged installed then you can convert your ext3 to use htree's .. Htree's are basically an indexed directory support for the ext3 filesytem, it offers a considerable performance gain when used on a filesystem with directories containing a large numbers of files ( hopefully I didnt confuse you there ). Now if you know for a fact your kernel is up to date and supports htree's ( if you dont know then DO NOT try this, don't be stupid and assume it will work ) then issue the following to make sure your e2fsprogs is at least 1.32
e2fsck -v , if its up to date ( at says its at least 1.32 ) , then issue this to add htree support to ext3
tune2fs -O dir_index /dev/hdXXX
Voila your done, I recommend you reboot after you do this.. But you don't have too. Now If you feel outdated and dissapointed b/c you couldn't add htree support to ext3, then don't worry htree ext3 users and non - htree ext3 users can still add the no access time tweaks and thus still speed up your ext3 or ext2 filesystem..
To add them basically edit /etc/fstab and follow how I change my examples
if your ext3 partiton looks simular to this.. /dev/hdb1 /tmp ext3 defaults 0 0
then your going to change it , to look simular to this /dev/hdb1 /tmp ext3 defaults,noatime,nodiratime 0 0
Now either reboot or remount your filesytem. Ok well that's enough for the filesytem .. Next on the list is changing pci latency , setting up ntp ( for syncing time , not really a tweak but its useful ) and then finally kernel tweaks ( basically the do's and don't's for compiling a 2.4 or 2.6 kernel , yes that includes how to setup 1000hz amongst other things )
Matt -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.2.3 (GNU/Linux)
iD8DBQE/T95ZleY/n9G/oZ8RAovXAJ9t1aeNc9Kj75SYT9x8Sk8QWNXBegCff8wk ulEb1LMgU4nzgXsiX92bMH4= =okTZ -----END PGP SIGNATURE-----
_______________________________________________ To unsubscribe, edit your list preferences, or view the list archives, please
http://list.valvesoftware.com/mailman/listinfo/hlds_linux
_______________________________________________ To unsubscribe, edit your list preferences, or view the list archives, please visit: http://list.valvesoftware.com/mailman/listinfo/hlds_linux
_______________________________________________ To unsubscribe, edit your list preferences, or view the list archives, please visit: http://list.valvesoftware.com/mailman/listinfo/hlds_linux
