I've noticed this in the past on one of our boxes and am hoping for a
bit of validation before I start chasing wild geese round the kernel.
This is an older Alpha running 3.2C and it has servers for HTTP, NNTP
and RADIUS on it. A few users telnet into it to run some apps, but a
significant delay is noticed during login ... such that some have not
been able to login reliably. The odd thing about this is that we are
seeing the majority of the connection failures from internal users at
a remote site (which I would guess indicates a routing problem), with
these same users having no difficulty going to outside hosts or other
systems on the same segment as this problem Unix box.
Given the number of sockets tied up in *_WAIT states when I 'netstat'
this box -and- the bursty nature of WWW serving, I'm guessing that we
need to do some tuning here ... but that doesn't explain the apparent
"localization" of the problem. To make matters worse, it seems to be
a consistent problem with all of our Digital Unix boxes.
Does this ring any bells with anyone? I'm a bit puzzled and I *know*
I'm not providing enough information, but am looking for any place to
start working on this mess.
andrew. (brennan_at_auhs.edu)
"Last laugh? I don't want it -- laughter wasn't meant to cease."
Received on Tue Oct 13 1998 - 23:08:05 NZDT