LANCP counters

[Art Wiens]

I've been trying to isolate some networking issues and was wondering how much I can "trust" the counters displayed with a SHOW DEVICE xxxx /COUNTERS ?

They "seem" legitimate on some Alpha's (VMS v7.2-2) but are very suspicious on some VAX's (VMS v6.2). There is a LANACP process running on these systems.

There doesn't seem to be a command to ZERO the counters which would be handy (assuming the counters are accurate). Is there a way to do so other than reboot?

Art

[Art Wiens]

BTW, the counts in the attachment are from two systems that have been up for 108 days.

Art

[Galen Tackett]

Art,

In my experience, too, these counters have been quite reliable on the Alpha. It's too far back for me to remember such experience from Vax usage, but I wonder if they might be 16 bit counters on the Vax implementation?

I don't believe there is a way to clear these counters short of maybe re-initializing the device with the appropriate $QIO function code. But that would mean shutting down whatever was using it first.

If you are running DECnet Phase IV you can use NCP like this:

$ RUN SYS$SYSTEM:NCP
NCP>SHOW LINE EWA-0 COUNTERS ! To see the counters for EWA0:
NCP>ZERO LINE EWA-0 COUNTERS ! To zero them
NCP>SET LINE EWA-0 COUNTER TIMER 1000 ! To zero them every 1000 secs

To make the counters always reset every 1000 seconds even after you restart DECnet or reboot use:

NCP>DEFINE LINE EWA-0 COUNTER TIMER 1000

Hope this helps.

--
Galen

P.S. I have some code that lets you monitor counters over the LAN provided that the device speaks MOP. All the DEC heritage devices I've ever tried do. I can't make it available any time in the near future but hope I can some day.

[Art Wiens]

Thanks, I'm aware of the Decnet line/circuit/node counters, but the majority of traffic is not Decnet so I was looking for some "higher level" counters. Majority of interactive users use KEA (Telnet) and file transfers are mainly FTP, although Decnet transfers are also done.

Art

[Andy Bustamante]

It sound like you may want to try $TCPIP SHOW PROTOCOL to focus on TCP/IP traffic.

[Uwe Zessin · Accepted solution]

Unless it has changed recently, the DECnet LINE and LANCP DEVICE counters are the same (modulo some minor differences...) - they come from the hardware interface.

Tried it on V6.2-1H3 and a:
NCP> ZERO KNOWN LINES COUNTERS

did reset the numbers in LANCP's display, too.

[Uwe Zessin]

Something that has just popped up from the depth of the grey mass ;-)

How is the VAX connected to the network? If I recall correctly, 'Collision detect check failure' can be a symptom of a bad 'heartbeat' setting on those mini transceivers.

[Art Wiens]

Andy:

You mean NETCU SHOW COUNT (We're a TCPware env), but yes thanks. I'm trying to correlate what they report to what LANCP shows.

Uwe 1):

Really?! So theoretically if I zero the Decnet line counters, I should see that reflected in the LANCP counters? Lets try that...

Uwe 2):

The VAX's are connected with transceivers and cat 5. I've been doing some reading about Collision Detect circuitry and how the switch pack is set on the transceiver... it doesn't appear to matter if these systems are on a network switch, only if they're on a hub. If they're set the "wrong" way on a switch it increases the count of Collision Detect Check "Failures" but it's not really a problem.

Art

[Art Wiens]

Hey Uwe, reply once more...you shall receive many points for your first tip!! See attached.

So then if that's the case, IP counters are over and above what LANCP/Decnet line counters are showing? Is that because we're using TCPware and not "native" TCPIP ?

Thanks,
Art

[Jess Goodman]

On VAX VMS 6.2, and on Alpha VMS 6.2 without ECO ALPLAN06_062 installed, the LANCP counters are the same 32bit counters that NCP uses, so NCP ZERO KNOWN LINE COUNTERS will zero them both.

But I would *not* trust the LANCP SHOW DEVICE /COUNTERS display on these system because there is no indication when the numbers overflow 32 bits, unlike the NCP SHOW KNOWN LINE COUNTERS display which use a ">" before numbers that have overflowed. At our site the bytes sent/received counters can overflow in less than an hour.

On VMS 7.2+ Alpha (I don't know about VMS 7.0, 7.1, or VAX 7.2) the LANCP counters are separate from the NCP counters and use 64 bit cells so overflow isn't a problem. On these systems the command to zero them is (from FAQ):

LANCP SET DEVICE/ALL/DEVICE=FUNCTION="CCOU"

On Alpha VMS 6.2 with ECO ALPLAN06_062 installed the LANCP counters are 64 bit, but there is apparently no way to zero them out.