[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: performance problems
On Tue, 24 Nov 1998, Erez Doron wrote:
> Leandro Guimaraens Faria Corcete Dutra wrote:
> >
> > Erez Doron wrote:
> > >
> > > > > because processes are starting and ending all the time,
> > > >
> > > > You could run all these programs as daemons from the
> > > >
> > > all thease programs are daemons, which ran from inetd !
> > > ( like pop, ftp,telnet etc...) i can not run them from /etc/rc...
> >
> > Well then, they are not starting and stopping all the time! Daemons
> > may fork now and then according to load, but they will neither start nor
>
> inetd forking and then execve ipop3d, is same as starting ipop3d !
>
> > stop unless the machine reboots or the superuser kills them.
> >
> > Have you tried already running top in a small window and checking which
> > process is "eating" you CPU away when the cursor halts? I think I have
>
> well, both cpus are below 10% usage, so no process is eating my cpu
>
> but I think that the problem is syslogd, which does not write to
> cache, but directly to disk, so if it has somthing to write, it freezes
> the system untill it's done.
>
> i soved the problem by killing syslogd, but that's not a good idea
> > no other ideas...
Here is one: mount a ramdisk and arrange for syslogd to write there. You
will have to be careful not to fill the partition (only 4M w/o hacks).
This is MUCH faster than HDD. You can mirror it to HDD once in a while
using a cron script or a small program. The other idea is to stripe the
/var/adm so it's on a different drive than most other things. The third
idea is to make syslogd log on another machine ONLY (see syslog.conf).
There are also other performance tuning issues for machines that run a
lot of processes, such as slowing down the task swapper in the kernel and
changing the swapper and cache buffer page policy. These cannot be done
easily by a user/sysadmin (I think you can choose the swapper policy at
compile time by selecting 'optimize as server' in make config - but don't
take my word for it). There is more to be done, such as setting the
interrupt priorities differently and looking into real time Linux, which
can guarantee latencies to some extent.
Erez, you must be one of the few people in the world who are stressing
Linux to the limit all the time ;) No-one that I know runs a production
network from a machine that is used as an X terminal/server at the same
time ;) And I think that it's funny that the only complaint is, slow
cursor response in X (which may be due to the mouse interrupt having a low
priority vs. disk and ether).
Peter