bandwidth usage in bytes on Linux 🔗 link

Jesper Johnson wrote:

Hi

I'm looking for a way to create graphs for hobbit on Linux to monitor network bandwidth usage, separately on each interface (eth0,eth1). The provided netstat rrd Network I/O graph measures packets/second but I want bytes/second, which netstat0 (as defined in hobbitgraph.cfg) is supposed to graph but it doesn't since the netstat -s command on my server (RHEL4) doesn't return any byte statistics (only packets).

I suppose I could build something like this (monitoring the /proc/net/dev file) myself if I looked into setting up a custom rrd and so forth, but I'm hoping to save some trouble by asking if someone here has already done this :-D

Help, anyone?

-- 

On Thu, Nov 10, 2005 at 06:46:49AM -0600, Rich Smrcina wrote:

Jesper Johnson wrote:

I'm looking for a way to create graphs for hobbit on Linux to monitor >network bandwidth usage, separately on each interface (eth0,eth1). The 

There is a very well described MRTG interface for Hobbit.  I just implemented it at a customer site and it works as advertised.  The URL is in the help.

I dont think Hobbit should use these to populate the tcpOutDataBytes and
tcpInInorderBytes fields in the netstat rrd - the ifconfig counters are
per network interface, where the rest of the netstat data are global
counters. 
But perhaps there should be a completely separate set of RRD files, with one graph per network interface ?

Henrik Stoerner wrote:

On Thu, Nov 10, 2005 at 06:46:49AM -0600, Rich Smrcina wrote:

Jesper Johnson wrote:

I'm looking for a way to create graphs for hobbit on Linux to monitor network bandwidth usage, separately on each interface (eth0,eth1). The 

There is a very well described MRTG interface for Hobbit.  I just implemented it at a customer site and it works as advertised.  The URL is in the help.

MRTG would indeed solve this problem.

But it might still make some sense to have the Linux hobbitclient collect the data from /sbin/ifconfig. It does have byte counters:

eth0      Link encap:Ethernet  HWaddr 00:0E:A6:CE:D6:85            inet addr:172.16.10.100  Bcast:172.16.10.255  Mask:255.255.255.0
          inet6 addr: fe80::20e:a6ff:fece:d685/64 Scope:Link
          UP BROADCAST RUNNING MULTICAST  MTU:1500  Metric:1
          RX packets:7935102 errors:0 dropped:0 overruns:0 frame:0
          TX packets:7943900 errors:0 dropped:0 overruns:0 carrier:0
          collisions:0 txqueuelen:1000           RX bytes:2210282597 (2.0 GiB)  TX bytes:2268014990 (2.1 GiB)
          Interrupt:17 
I dont think Hobbit should use these to populate the tcpOutDataBytes and
tcpInInorderBytes fields in the netstat rrd - the ifconfig counters are
per network interface, where the rest of the netstat data are global
counters. 
But perhaps there should be a completely separate set of RRD files, with one graph per network interface ?


Regards,
Henrik

-- 
Rich Smrcina
VM Assist, Inc.
Main: (262)392-2026
Cell: (XXX)XXX-XXXX
Ans Service:  (360)715-2467
user-61add9955ef9@xymon.invalid

Catch the WAVV!  http://www.wavv.org
WAVV 2006 - Chattanooga, TN - April 7-11, 2006

On Thu, Nov 10, 2005 at 09:14:43AM -0500, Scott Walters wrote:

But it might still make some sense to have the Linux hobbitclient
collect the data from /sbin/ifconfig. It does have byte counters:

eth0      Link encap:Ethernet  HWaddr 00:0E:A6:CE:D6:85
          RX bytes:2210282597 (2.0 GiB)  TX bytes:2268014990 (2.1 GiB)

I dont think Hobbit should use these to populate the tcpOutDataBytes and
tcpInInorderBytes fields in the netstat rrd - the ifconfig counters are
per network interface, where the rest of the netstat data are global
counters.

Correct, and I've never really trusted those counters for 'real bandwidth
usage'.  I think re-transmissions and other TCP counters need to be added.

But perhaps there should be a completely separate set of RRD files, with
one graph per network interface ?

Yes.  And for the other UNIXes you'll want to to run these local commands
and send them to the server for parsing:

AIX|HP-UX: /usr/bin/netstat -v

Solaris: /usr/bin/kstat -p -s [or]bytes64

And a 'netstat -i' provides ethernet layer stats that can be useful.  Not
for bandwidth, but Input Errors and Output Errors are never a good thing.

Henrik, do you need sample outputs to do the parsing on the server?

Henrik Stoerner wrote:

But perhaps there should be a completely separate set of RRD files, with one graph per network interface ?

On Thu, Nov 10, 2005 at 09:20:56PM +0200, Jesper Johnson wrote:

Is implementing this into hobbit a lot of work or is there a possibility 
we'll be enjoying this new feature before 2006? :-)

Last but not least a big thank you to Henrik for hobbit. Found out about 
this project a few days ago... it's great!

Henrik Stoerner wrote:

I took a shot at it tonight, and I think I have something working for
linux boxes. I'll let it run overnight and see if the data look right in
the morning. I need to do the parsing of the other OS's output formats,
but that shouldn't be much of a problem. Regular expressions are great.

On Tue, Nov 22, 2005 at 02:33:00PM +0200, Jesper Johnson wrote:

How did your 'bandwidth graph per interface' -experiment on Linux go and is 
there a test version available somewhere yet?

Henrik Stoerner wrote:

It actually seems to work quite well. It is currently only for Linux,
but it has been included in the nightly snapshots for some time - 
you'll find it at http://www.hswn.dk/beta/

The clients must be updated, so they report the raw "ifconfig" data
(for Linux - there are some other commands for AIX, Solaris and HP-UX,
 but I haven't done the server-side parsing for these OS's yet).