stale telnetd procs

[Bill Warner]

#uptime
  8:51am  up 20 days,  9:52, 511 users,  load average:  8.04, 6.56, 5.67

17 days isn't much yet but this is 5pm after most people have started
leaving.
this server runs 550-650 users all day with load of 5-30 being average.
this
is concurrent telnet sessions.  some programers some sales some admins
...etc...

We are having a big problem with a kludge to work around it.  It seems
that when
people log out telnetd doesn't let go of the connection.  Before out
kludge we were
able to drive the load to over 700!!!  great thing was that the box just
chuged allong
slow but stable.  (the sco box it replaced would die if  load ever went
over 90).  
I have found some /proc settings that are suppose to help this.  mainly:

/proc/sys/net/ipv4/tcp_keepalive_time

witch defaults to something like 3 hours.  I dropped this to 10 minutes
and we still
are having the same problem with telnetd not releasing the connection. I
have added
all these to help performace into /etc/rc.d/rc.local

#Network preformace
echo 30 > /proc/sys/net/ipv4/tcp_fin_timeout
echo 600 > /proc/sys/net/ipv4/tcp_keepalive_time
echo 0 > /proc/sys/net/ipv4/tcp_window_scaling
echo 0 > /proc/sys/net/ipv4/tcp_sack
echo 0 > /proc/sys/net/ipv4/tcp_timestamps

Finally we made a simple ps | grep <stuff> script that would find the
stale telnetd procs and
kill -9 them.  This script has to run ever 7 minutes in order to keep
the load down.  Is there 
any other settings I can look for that might help with this problem?

PS does anyone else use linux with this many telnet sessions in all atl
at the same time?



-- 
--
Bill Warner
Direct Alliance Corp.
Unix/Linux Admin.