[K12OSN] OT: still have sluggish system

[dahopkins at comcast.net]

It has been a long day.

> Turn off nscd. Add entries in /etc/hosts for the authentication
server(s). 

Done, did not make a difference 

> Check /etc/nsswitch.conf for lookup ordering on DNS and
authentication. DNS lookup should be set to "hosts: files dns" in
nsswitch.

Done, the above were the current settings. 

> Check "/sbin/chkconfig --list"  for a list of all services that are on
at boot up. Turn off everything that is not used with "/sbin/chkconfig
<service name> off"

Done , turned off everything I could reasonably think would not be needed (including iptables, bluetooth, etc) 

>Check in /etc for multiple nsswitch.conf files. Be sure the correct one
>is in use. Scan for rpmold and rpmnew.

Done, only have one nsswitch.conf.  Not sure what rpmold/rpmnew are.  But .. system is fully patched.

> Then reboot the machine(s) and go to the closet and unplug the client
>connection to the switch. See if the load spikes on the server. If it's
>OK, start plugging in segments to see if the switch is bogging down and
>if so, on what segment. Keep an eye on the server load.

It is some form of broadcast traffic on the system.  Since I have cups running, and also 2 Windows TS, and samba, I guess this might be expected, just seems somewhat more than typical.  

>Also check the /etc directory for multiple dhcpd.conf files. K12LTSP v.5
>uses a differently named file than dhcpd.conf.

Using the dhcpd-k12ltsp.conf file for configuration.  Not sure how Eric set up dhcp to use this, but ... it works.

>Lastly, check the clients to see what video driver they are running. If
>they are using the old vesa driver, that is a hog (painfully slow) and
>could be a primary detractor. K12LTSP v.5 now uses xorg and not xfree.
>Some older stuff has been dropped.

All the clients are based on the Via Epia 533 chipset.  

I have noticed with ntop that a alot of thin clients show up as multihomed.  Not sure why since they only have one entry in the dhcp-k12ltsp.conf file.   Also, some of them have rather high values for NFS(???),  but this is not specified anywhere in the lts.conf file.  The only other change was using glx, I have disabled that as well.

Selinux is not enabled.  I have used all of the suggestions from all of the responses received.   Also, did ethtool -K eth0 tso off as well.  There was a post from early July on another list that watchdog may be an issue, but ... not disabling that I think. 

So ... I am left with no idea what is going on.  However, the Tech Coord and I have been here since 7pm (it is now 2am) testing various options with rdesktop and using our backup K12LTSP server as well (running FC4 release).   Not sure what we have actually ruled out :( but at least we know the behavior is the same with FC5 and the prior version at this point.  

The basic issue is that for the last 3 years, we have been using Type-to-Learn via rdesktop to our student terminal server very successfully.  Now, after upgrading, once we reach about 9 users, the Windows Terminal server hits 100% CPU  and everything comes to a halt.  Previously, we were running without issues with 30 users (and as many as 40 on occasion), server was still reaching 100%, but it worked.  The windows systems  are also fully patched.  

So, in about 6 hours, the kids start arriving again, and ...we still can't use TTL.  The tech teacher is ready to scream, as am I.  Not sure what my next step is at this point.   I could rebuild the windows system but what will that buy me.  The only thing I can think of is that either a hotfix changed something on the windows system, or ... perhaps the most recent version of rdesktop may be the issue (it was release last March?) or ... I need to revert to the 2.4 kernel for the thin clients (just thinking/musing out loud).  Oh, well, time to go get about 3 hours of sleep.

Again, thanks ... I just hope this can be resolved soon, before the parents start to get upset.

Sincerely,
Dave Hopkins