Slow/no response
Ian Carr-de Avelon
avelon at emit.pl
Fri Oct 22 08:22:04 UTC 1999
Ralf Hildebrandt wrote:
>On Thu, Oct 21, 1999 at 04:21:30PM +0100, Ian Carr-de Avelon wrote:
>> I have a problem that looking up hosts on our system occasionally becomes
>> slow and then searches start to fail. This even affects local zones. The
>> problem occures at times of high congestion on the net, which leads me to
>
>If the net is congested, how should BIND's answers come through?
I am obviously not being clear enough here. I don't mean that my own
net is congested, or my link to the Internet, I mean that the international
links are congested. The inability to resolve is exactly the same on the
name server itself as on the rest of the local net. I would expect that
distant zones would resolve progresively slower and then start to time out.
I would expect resolutions from cached zones to remain fast (obviously
it is difficult to be sure that cached information has not exceeded its
time to live. I would expect that resolution of zones for which the
server is the primary would always remain fast. So if I type:
host www.emit.pl
on the server and that never resolves, something has gone wrong. This
happens after named has been running for hours/days without problems. If I
kill named, I hear a large amount of disk activity for a few seconds
and a newly started named will resolve local addresses as expected
for a few minutes, before the same problem occures.
S Trefethen wrote:
>Just a thought, but you might want to check to see if your server is
>swapping. *ANY* swapping can slow dns down considerably, lots of swapping
>will severely impair a dns box.
What I see on the system is:
CPU states: 0.7% user, 0.3% system, 0.0% nice, 99.1% idle
Mem: 30716K av, 27404K used, 3312K free, 6460K shrd, 2304K buff
Swap: 142408K av, 5400K used, 137008K free 12312K cached
PID USER PRI NI SIZE RSS SHARE STAT LIB %CPU %MEM TIME COMMAND
846 Web 3 0 7080 6528 1076 S 0 0.0 21.2 32:51 roxen
12957 root 1 0 2108 1460 412 S 0 0.1 4.7 0:21 named
Obviously there is a very little bit of swap used there, but that seams to be
typical of the way Linux manages memory. I don't hear any thrashing of the
disk. I can maybe play with setting named to a negative nice value, but
it is not that the program really stops. Strace shows that there is a
sort of heartbeat going on, but it never answers the resolver.
Yours
Ian
More information about the bind-users
mailing list