network resources

[Stuart Young]

At 08:58 AM 19/12/00 -0500, Charles Crawford wrote:
>Hi,
>
>I'm having problems getting Samba shares to respond quickly enough to not
>time-out on our NT network.
>
>I've got a login script that creates a mapped drive to a share, but it tells
>me that the server is not responding and that I might not have enough
>network resources available to make the connection.

What 'Security' option are you using? Who is verifying your users passwords?
Sounds like you are using 'Security = server'. Are you sure the issue isn't 
the 'server' (which I would guess is an NT box) is taking a long time to 
process password requests?

>The systems that run Samba are all Red Hat Linux 6.1 systems, with Samba
>2.0.5a-12 on one, and Samba 2.0.7-4 on the other. I don't think that this is
>a Samba issue directly, but it is causing some difficulty and is causing
>concern among management in regards to the effective use of Samba in our
>networking environment.

How are you running Samba? I'd assume you are running it as a daemon, but 
if you are running it from inetd, then startup will be VERY SLOW.

If you are running it from inetd (which I seriously doubt if it's installed 
as rpm's), then it is probably also being called through tcp_wrappers 
(/usr/sbin/tcpd) which usually does a name lookup on every connection. If 
you resolver has problems, this will cause a delay at start, at least for 
the first connection (after which the delay will happen again once the 'bad 
response' has timed out, and then again, and again.. *sigh*).

Also what sort of CPU load is on these boxes?

>Once the connection is made, which sometimes takes several login attempts,
>there appear to be no problems. One item worth noting, however, is that
>pinging any of the Linux machines brings a response time 2 times that of the
>Windows machines. I've made sure that the IP addresses are included in the
>WINS database, the host and lmhost files, and in the Server Manager.

What machine are you pinging from? How is the network laid out?

>The names resolve to the correct machines, but sometimes the connection is
>very slow. I'm starting to think that Linux is misconfigured to allow a
>small number of network connections, or that the switches are bottle-necking
>the network connections. Anyone have any ideas?

How long does the resolve name issue take? Do you really need the machines 
names to resolve under the Linux side? If not, there are various things you 
can do, even within Samba to disable the use of DNS. You could try removing 
the 'bind' option from the order statement in the /etc/host.conf file. You 
could also try removing /etc/resolv.conf and try that out. Back up the 
files first of course! If this resolves your connect speed issue, then you 
definitely have a name lookup problem. Either disable name lookups thru 
Samba or on the Linux boxes, or alternatively run an internal DNS that will 
resolve internal IP's and forward external requests to your external DNS.

A last possibility that I can see (and doubtful) is that your network is 
having lots of browser elections. You could try using a network sniffer 
like Ethereal, or possibly using an more generic tool like IPTraf to look 
for such traffic. If it seems excessive, you should probably remove a lot 
of your machines from performing in browser elections. (Anywhere you find 
'browser' in the smb.conf man page is probably thing to good look at, but 
'preferred master', 'local master' and 'os level' are the main ones). This 
should actually cause more of a general slow-down, but YMMV.


Stuart Young - sgy at amc.com.au
(aka Cefiar) - cefiar1 at optushome.com.au

[All opinions expressed in the above message are my]
[own and not necessarily the views of my employer..]