[rt-users] Performance (speed) of RT3

[John Jasen]

On 19 Nov 2003, John Schubert wrote:

> On Thu, 2003-11-20 at 10:03, John Jasen wrote:
> > If this machine is only used for RT, and there are only 3 people using it, 
> > setting the max servers to something like 5 should be reasonable. You can 
> > also play with the MaxRequestsPerChild value until you get a comfortable 
> > balance between performance and memory usage from leakage (mine, on a 1GB 
> > system doing entirely too much else, is 500).
> I should have provided more information.  Currently 3 are beta testing,
> but ultimately 14 people will be logged in (creating tickets), with a
> dozen or so MORE users who may just view tickets.  So should I tweak
> maxservers to around 50?  

users + viewers + only a few spare servers for if something goes wonky.

You don't have enough ram to support 30 simultaneous users in the box, 
however. I think apache takes at leats 8 megs for each process ...

> > If you're using mysql as your database, mysql 4 (I'm not sure about 3) is 
> > faster if you use innodb tables, and you can tune it for how much system 
> > resources it can gobble up.
> At this risk of asking the obvious, is there a good source you know of
> for finding how to fine tune MySQL 4?  Googling provides too vast of an
> array (some are too basic, some too vague).

Try looking in Mysql 4 for my-$name.cnf, where $name is huge, large, 
medium and small, for a start. Read those, and you start getting a basic 
idea.

> It does help greatly.  I really appreciate the responses I've gotten
> from the list.   Unfortunately I have little experience tweaking, since
> previous experiences with *nix have been on Ultra10s with ample
> resources for everything I threw at it (apache, NIS+ slave, MySQL, &
> some others).  I bought a Linux book, but it doesn't address fine tuning
> at all.  Googling only had some vague references to RedHat.com with
> ambiguous information, such as use Top.  Garbage in, garbage out, so I'm
> sure I'm to blame for not finding better information.  So I thank
> everyone on here for their indulgences/patience.

performance tuning is an art, not necessarily a science, and involves a 
lot of knob twiddling, testing, twiddling and testing, twiddling and 
testing ...

And then your requirements change. :(