From: alan.nguyen@au.transport.bombardier.com
Date: Sun Feb 16 2003 - 19:37:27 EST
Thank You to, Tom Webster, Jesper Frank Nemholt, Pat O'Brien, Paolo Lucente,
Dominic Christopher, Charles Ballowe, for all your helpful advices, info, and
hints especially Jesper, Tom and Charles.
The original post:
> I need to understand the basics of how to measure the Bandwidth Utilization of
> the WHOLE network (Ethernet and FDDI).
>
> I have a copy of mrtg, but this shows only the incoming and outcoming packets
> through that machine only , not the whole network performance.
> The network bandwidth is 100 Megabit/second, measured in bit/sec, but the
> netstat or collect tool show only the packets. So how many bytes in a packet ?
> How to convert to bit/sec ?
Helpful Info:
Depending on what you are actually trying to do and the physical layout
of your network, it's wither simple or near impossible.
The first thing to realize is that in a switched network, like most
modern Ethernet networks, it's impossible to see all of the traffic from
a given node.*
* Unless you do some really silly things that are going to mess with your
network performance.
What you can see is the aggregate throughput of the routers and switches
on your network. It may be enough to show the throughput on the router
that connects your Ethernet and FDDI networks is you are only looking
for link utilization between those networks.
If you are going to try to get an idea of how much traffic is on all parts
of your network, you are going to need to have every switch report data.
MRTG is the right tool for this, depending on the hardware and the
measurements supported by the SNMP MIBs. In pretty much all cases, you
should be able to get a measurement of the traffic going across the
backplane of the switches measured in (M/K/bits per second).
--------------------------------------------------------------
i think if you want stats for an entire network you have very little
chances to do so, via a Tru64 machine. Tipically if you want "numbers"
if could use accounting/snmp stats exported by switches/routers involved
in sending back and forth your packets on your network.
For example if you need some kind of statistics for your internet
connection bandwidth utilization you should use accounting/stats utils
common on most routers. If you need some kind of stats for your LAN/WAN
utilization, SNMP stats exported by switches (exactly as you do it with
a Tru64 machine) have a granularity of physical port connections: so if
you export and publish all your stats on a single collector you should
have the entire picture.
An alternative solution for a LAN of small-to-mid sizes could be to have
a single machine with a bunch of NICs, connected to a number of switches
that replicate all traffic on the port that computer is connected on; so
with some utility (not available for Tru64) and promiscuous mode you
will be able to get numbers and times in a very rough form. Than you
have only to process them to extract averages.
--------------------------------------------------------------
Ethernet and FDDI can be both shared, switched and with/without duplex.
On a shared network you'll need to monitor the hub that interconnect all
machines. Ofcourse it needs to be a hub supporting SNMP to do so. On a
shared network you _can_ simulate the same by putting the NIC in
promiscious mode and thus see all traffic on the segment, but I doubt
MRTG will allow to monitor it.
On a switched network you'll need to monitor the switch, and it'll
likely give statistics for each port plus a summary for all ports
(backbone).
---------------------------------------------------------------
It's a tough question to answer. It largely depends on network architecture.
Are you on a switched network? Is it just hubs? What kind of switches?
Double check the port settings on the switch and ensure they match you
system -- it can have huge impact on performance. Also, if you have snmp
capable switches, point mrtg at them. Pay close attention to the uplink
port on the one that connects to your systems. Is the slowness between
2 boxen on the same switch or only from separated points of the network.
Look for bottlenecks between the 2 endpoints. If mrtg isn't showing the
port on your server as maxed out, odds are you're going to need more digging
and a deeper understanding of your network architecture than I can guessa
at.
Alan.Nguyen@au.transport.bombardier.com
This archive was generated by hypermail 2.1.7 : Sat Apr 12 2008 - 10:49:08 EDT