topic Re: LAN Statistics in Operating System - HP-UX

LAN Statistics

EML — Tue, 12 Nov 2002 10:26:45 GMT

Would it be possible to have some statistics on a file, on the the use of lan cards on a server.

I have a server with 9 lan ports use and I want to monitor the performance of each lan. Something like a text file that may contain lines every 5 minutes to put into Excel and do graphs?

I have looked at lanadmin, but the statistics are on Menu mode. Do you know of any other tool I can use?

Re: LAN Statistics

Elif Gius — Tue, 12 Nov 2002 10:34:45 GMT

what's with:
netstat -in
there you get some information about incoming,outgoing packages...run it in a loop ...
while true
do
netstat -in >> /tmp/netstat.out;sleep 30
done

Re: LAN Statistics

Stefan Farrelly — Tue, 12 Nov 2002 10:36:25 GMT

You can use netstat -i for packet counts.

The best tool though is either lanadmin or glance/gpm.

You can use lanadmin in a script to collect stats by passing commands to it, eg;

lanadmin <lan
display
quit
EOF

Then grep for what you want - eg. input collisions or errors.

Re: LAN Statistics

Elif Gius — Tue, 12 Nov 2002 10:38:03 GMT

my colleague gave me tip:

on Linux you can use tcpdump to dump traffic on a network...it is also available for hp-ux
look at this link :

http://hpux.asknet.de/hppd/hpux/Networking/Admin/tcpdump-3.7.1/

Re: LAN Statistics

Steven Gillard_2 — Tue, 12 Nov 2002 10:43:43 GMT

Do you have the performance agent (measureware) running on the box? You can check with "mwa status". If so, it records the following metrics for each interface at 5 minute intervals:

**..................... Network Interface Summary Metrics

* BYNETIF_NAME
* BYNETIF_NET_SPEED
* BYNETIF_NET_MTU

**..................... Network Interface Network Metrics

* BYNETIF_IN_PACKET_RATE
* BYNETIF_IN_PACKET
* BYNETIF_OUT_PACKET_RATE
* BYNETIF_OUT_PACKET
* BYNETIF_IN_BYTE_RATE
* BYNETIF_IN_BYTE_RATE_CUM
* BYNETIF_OUT_BYTE_RATE
* BYNETIF_OUT_BYTE_RATE_CUM
* BYNETIF_ERROR_RATE
* BYNETIF_ERROR
* BYNETIF_COLLISION_RATE
* BYNETIF_COLLISION

Use the 'extract' command to dump this information into a flat file for importing into other tools. It has an interactive interface to guide you through the process if you're not familiar with it.

Regards,
Steve

Re: LAN Statistics

Chuck J — Tue, 12 Nov 2002 10:47:30 GMT

Use the netstat command with options. Some options that might be of interest are:

-i = state of all network interfaces
-n = show network address numbers
-s = show stats for all protocold
-a = show the state of all sockets

you can also get some graphical stats from GPM.

Chuck J

Re: LAN Statistics

Thayanidhi — Tue, 12 Nov 2002 10:49:02 GMT

Hi,

Try to use

lanadmin -g mibstats

Generally ppa is 0 for lan0 and 1 lan1 and so on.

use lanscan command to find out ppa numbers.

Note: The above is working for 11.x only.
For 10.x nmid is used instead of ppa. But I am not sure whether -g option available in 10.x
lanadmin

TT

Re: LAN Statistics

Ron Kinner — Tue, 12 Nov 2002 11:08:45 GMT

If you want graphs of traffic on your LANs get MRTG. It's free and runs on almost any platform and can collect data from every device in your network.

http://people.ee.ethz.ch/~oetiker/webtools/mrtg/

Ron

Re: LAN Statistics

T G Manikandan — Tue, 12 Nov 2002 11:42:42 GMT

You can use tcpdump/ethereal.
HP has a inbuilt utility for network tracing
Check the man pages of nettl.

check the document which will help in using nettl.

Re: LAN Statistics

rick jones — Wed, 13 Nov 2002 19:09:08 GMT

i would second the landadmin -g mibstats stuff - it removes the menu mode from older versoins of lanadmin. you do need to be reasonably up-to-date on patches.

mrtg would also work, assuming you are still running an SNMP daemon on the system. some people decided to turn those off.

the measureware stuff is goodness too...

however, for all of those there is something to keep in mind - all the byte/octet counters in all three cases are (iirc) still only 32-bit quantities. a 100BT interface running flat-out will wrap those counters in something just shy of 7 minutes. A gigabit interface not even running flat-out would wrap those counters in less, so you may need/want a sample interval of less than five minutes.

and of course, encourage the adoption of 64-bit counters for network statistics...