topic Re: TCPIP Pause At Startup in Operating System - OpenVMS

TCPIP Pause At Startup

Jack Trachtman — Fri, 14 Apr 2006 10:19:15 GMT

We sometimes isolate our VMS systems for
testing by physically disconnecting the
ethernet cable before booting.

I've noticed that TCPIP pauses near the
end of its startup and waits about 5
minutes before continuing. I've tried
inserting a turn-around ethernet plug
but there is still a pause.

1) What thing (or things) is TCPIP
startup trying to do at that point?

2) What timer(s) affect the above?

3) What's the simplest way to reduce
this pause?

Thanks all

Re: TCPIP Pause At Startup

Karl Rohwedder — Fri, 14 Apr 2006 10:55:40 GMT

The last thing TCPIP$STARTUP does, is calling your sitespecific startupprocedure (TCPIP$YSTARTUP) if available, before that it tries to register the network with SET NETWORK.

Are all service started wit 'normal' performance before that? May be something is waiting to connect to an outside server?

regards Kalle

Re: TCPIP Pause At Startup

Ian Miller. — Fri, 14 Apr 2006 11:31:31 GMT

do you have a dns server defined?
Parhaps there is a DNS lookup being performed.

Re: TCPIP Pause At Startup

Jim_McKinney — Fri, 14 Apr 2006 12:19:16 GMT

> What thing (or things) is TCPIP
> startup trying to do at that point?

Perhaps - attempting to establish a connection to your default router and update the routing table?

Re: TCPIP Pause At Startup

Volker Halle — Fri, 14 Apr 2006 12:33:59 GMT

Jack,

what you could try to isolate this, is:

Edit SYSTARTUP_VMS.COM:

$ SET PREFIX "(!5%T) "
$ SET VERIFY
$ @SYS$STARTUP:TCPIP$STARTUP.COM/OUT=TCPIP_STARTUP.LOG
$ SET NOVERIFY

and every DCL command line will be shown prefixed with the time in the .LOG file.

Volker.

Re: TCPIP Pause At Startup

Wim Van den Wyngaert — Tue, 18 Apr 2006 02:10:36 GMT

I just rebooted a VMS 7.3 station with TCP 5.3 ECO 2 and this with the network disconnected. Startup of TCP took 30 seconds (12 services enabled, no DNS).

It must be something specific for your site (NFS ? Setting in sysconfigtab.dat ?).

Wim

Re: TCPIP Pause At Startup

David B Sneddon — Tue, 18 Apr 2006 02:24:42 GMT

Jack,

Are you using SMTP? I have noticed a similar
issue on some of my systems (my wait is about
8 minutes) and I tracked it down to starting
the SMTP queues... don't know exactly what it
was doing but that is where my systems were
waiting.

Dave

Re: TCPIP Pause At Startup

Wim Van den Wyngaert — Tue, 18 Apr 2006 03:24:52 GMT

Just tested it with SMTP enabled. Still about 30 seconds.

Wim

Re: TCPIP Pause At Startup

Robert Atkinson — Tue, 18 Apr 2006 03:36:25 GMT

This is almost definitely a DNS problem, possibly SMTP trying to start or could be the LPD startup trying to translate the queue names.

We made this mod to our DR test plan, because effectively 'the cable is unplugged' and a reboot can take 10 minutes (as you said) :-

"13.10. Name Servers
In a test situation, or live invocation only one Domain Controller would be available, probably CODC-3.

The IP stack on the Alphaâ s is configured to look at all Domain Controllers, which means the time to return information is extended because of the timeouts required.

To ensure the lookups are as fast as possible, any DNS servers that are not operational should be removed from the list.

The list can be displayed by typing â $ TCPIP SHOW NAMEâ , and looking at the â serversâ field.

To change this, type â $ TCPIP SET NAME /NOSERVER=server_to_removeâ and â $ TCPIP SET CONFIG NAME /NOSERVER=server_to_removeâ , to make the change permanent."

Rob.

Re: TCPIP Pause At Startup

Robert Gezelter — Tue, 18 Apr 2006 04:39:21 GMT

Jack,

I would not recommend speculating as to what is happening. The science to determine the correct answer is not particularly complex. In addition, the same external behavior can be caused by many different actual problems (a problem that physicians are all too aware of).

The way to identify what communication is being attempted is to use a network monitor on the otherwise unconnected line to determine what IS happening, not what might be happening.

This used to be a question of non-trivial budgets, but tools like Ethereal ( http://www.ethereal.com ) are zero cost.

Capture the traffic emnating from the subject node with a sniffer and you will know precisely what the node is attempting.

- Bob Gezelter, http://www.rlgsc.com

Re: TCPIP Pause At Startup

Joseph Huber_1 — Tue, 18 Apr 2006 04:58:12 GMT

Whatever the reasons and methods to find out, if You just want to avoid the "pause" at TCPIP startup in this test boot situation, I would recommend to indicate through one of the USER* system parameters that no network is connected. This can then be used by systartup_vms to skip TCPIP (and related) startup.
If You disconnect the cable, then I assume You have access to the console and can do a conversational bootstrap to set and reset the system parameters.

Re: TCPIP Pause At Startup

Jack Trachtman — Wed, 19 Apr 2006 15:57:21 GMT

Thanks everyone for your thoughts.

Within the next week I should be able to try some tests. I will enable VERIFICATION
for TCPIP startup as one test.

In ref to ethernet sniffing - any way to use TCPTRACE from the host I'm booting? I realize this is like a chicken&egg (ie why should TCPTRACE be available before TCPIP startup completes?), but with the host off of the network, I can't run another box to sniff it.

Re: TCPIP Pause At Startup

Volker Halle — Thu, 20 Apr 2006 01:53:51 GMT

Jack,

you might be able to use tcpdump to trace all IP packets into (should be none if the NIC is disconnected) and out of your node.

$ @SYS$MANAGER:TCPIP$DEFINE_COMMANDS
$ SPAWN/NOWAIT tcpdump

TCPTRACE might also work, but you need to send output to a file and specify a high number for /PACKETS.

Try both methods manually first.

You certainly can't start tcpdump or tcptrace, before TCPIP is up, but you need to add those trace start commands into some of the TCPIP startup procedure ONCE you've figured out, where it hangs... Then add the start trace command in some procedure executed BEFORE that hang.

Volker.

Re: TCPIP Pause At Startup

Robert Atkinson — Thu, 20 Apr 2006 03:05:33 GMT

Jack - packet sniffing is a complete overkill.

As you can see from the example below, you'll get enough information from OPA0 to determine which module is giving problems :-

%%%%%%%%%%% OPCOM 8-NOV-2005 16:39:20.64 %%%%%%%%%%%
Message from user INTERnet on ALPHA
INTERnet ACP Activate RSH Server

%%%%%%%%%%% OPCOM 8-NOV-2005 16:39:51.56 %%%%%%%%%%% (from node BETA at
8-NOV-2005 16:39:51.24)
Message from user SYSTEM on BETA
%SMHANDLER-S-STARTUP, server management event handler startup

%%%%%%%%%%% OPCOM 8-NOV-2005 16:48:48.30 %%%%%%%%%%%
Message from user INTERnet on ALPHA
INTERnet ACP Activate SMTP Server

%TCPIP-S-STARTDONE, TCPIP$SMTP startup completed
%%%%%%%%%%% OPCOM 8-NOV-2005 16:48:48.72 %%%%%%%%%%%
Message from user INTERnet on ALPHA
INTERnet ACP Activate SNMP Server

You then need to work out why, which is almost certainly DNS related for these sort of delays.

What you need to start thinking about is how you can change the settings, i.e. remove unknown DNS servers before you boot the system.

Perhaps you could store a default version of the config file, and rename this over when you unplug your NIC cable?

Rob.

Re: TCPIP Pause At Startup

Wim Van den Wyngaert — Wed, 27 Jun 2007 07:35:03 GMT

I have now DNS enabled on my node.
I booted with no networks.
In hang for 3min30.

Jack : did you find the solution ?

Wim

Re: TCPIP Pause At Startup

Robert Atkinson — Wed, 27 Jun 2007 07:37:40 GMT

Wilm, what was the last console message before the boot hung?

Robert.

Re: TCPIP Pause At Startup

Wim Van den Wyngaert — Wed, 27 Jun 2007 08:01:27 GMT

The last message was "LPD symbiont has started".

I redid the boot with a tcptrace to sys$output. But it didn't work because you can not start tcptrace before tcpip.

So I did my interactive trace during startup of tcpip (trace for protocol IP). Not 1 record was shown. At the end of the startup (thus after the hang, records with to=from=127.0.0.1 were shown.

Wim

Re: TCPIP Pause At Startup

Wim Van den Wyngaert — Wed, 27 Jun 2007 08:02:42 GMT

The 127 records were snmp related. And only RCV.

Wim

Re: TCPIP Pause At Startup

Robert Atkinson — Wed, 27 Jun 2007 08:08:32 GMT

If you run '$ tcpip generate service /configuration/file=a.a' and have a look at the file, I'd bet the next service after LPD is SMTP?

Rob.

Re: TCPIP Pause At Startup

Wim Van den Wyngaert — Wed, 27 Jun 2007 08:14:34 GMT

Wrong. I did debugging and found that until the startup of rsh everything went smooth. It's the startup of SMTP that is delaying everything.

Wim