[Ref: Network Flow Analysis.]
Table of Contents
NetFlow is a network protocol developed by Cisco Systems for collecting IP traffic information. NetFlow has become an industry standard for traffic monitoring and is supported by platforms other than Cisco IOS and NXOS such as Juniper routers, Enterasys Switches, Linux, FreeBSD, NetBSD and OpenBSD.
A related protocol not discussed here, but is apparently supported by the described tools is: sFlow
Netflow gives us a closer view of traffic behaviour across supported network devices. The challenge is deploying tools to use Netflow effectively: collect the data, analyse, and summarise.
Manage a computer network and you need monitoring for an overview of traffic behaviour across your network. Manage an expanded network spanning a large number of hosts, devices, and/or geographical layout or multiple sites, and you really need network traffic monitoring/analysis tools. When you’re responsible for the network, everyone blames performance issues on your network and you need better knowledge of the security, stability and performance factors of your network. You need live data, and historical data with enough granularity to be able to extract useable information.
First look at Michael W. Lucas' book Network Flow Analysis. These notes augment the tool installation instructions from that book, where the human factor is important, in customisation/localisation, interpretation, we don’t do any of that here. Buy the book.
These notes are what we’ve done to successfully deploy netflow tools with OpenBSD. You will find other exorbitantly priced or free tools for monitoring the performance of your network, but Michael’s book gives invaluable examples for installing, validating and understanding your new netflow monitoring system
What netflow offers us, as shown in the diagram, is the ability to get network flow traffic from our routers/gateways (sensors), archive that data (collector.) With the data collected we can analyse the data either in ‘live’ data stream or from the archives.
Network Flow Analysis is an investment in your time. If you’ve got more time then sense, go ahead and try to get Open Source tools together and augment the information out there for how to do it.
We push data out of the router/gateways using Sensors/Netflow exporter using a directed UDP broadcast (i.e. data gets pushed in one direction to one target host.)
As indicated by the diagram, netflow export facilities (i.e. sensors) can exist in various devices. On gateway/routing appliances, such as from Cisco and Juniper, the sensor is often found pre-installed. Various software tools exist for Unix, including softflowd which is available in the OpenBSD Package Collection.
For these exercies in OpenBSD, we [document the use] of(netflow/sensor.html) pflow(4) deployed in the base install.
In its simplest form, the collector dumps the data to storage, at which point the data is available to your analysis tools.
The point of collecting and monitoring is to interpret that data so we can make informed decisions, take action. Data analysis can be on a separate terminal from the Collector, but for our needs we will use the same host.
Our common workflow is to monitor traffic latency using GUI tools such as smokeping, and CUFlow. When latency and throughput indicators from the above charts show dramatic changes, we seek more information (a more granular view) using FlowViewer.
Another workflow is to use FlowViewer/FlowTracker to track network traffic for specific services (for example, monitor traffic in and out of the primary file server or SMTP through the Internet Gateway.) When the traffic is noticeably averaging above a useability threshold, we have this collected data as justification for increase in either host capacity or link capacity for that host.
We describe here, three toolsets for analysis of the netflow data:
[Ref: Flow - Console]
The most powerful analysis tool is from the command-line using flow-tools
The best tutorial I’ve found is in print with Michael W. Lucas' book Network Flow Analysis.. We give you a sufficient start to make sure the data collection process is working correctly, together with your tool install.
If you’re networking kung fu is good, then use the command-line tools, but get the book and look at the following GUI tools.
We review the installation of 3 GUI front-ends to captured data. As with our above described workflows, a general overview of traffic behaviour can be extracted using our first tool: FlowScan/CUFlow.
To get more specifics on the traffic, and behaviour of specific hosts or ports, we then drill deeper using FlowViewer
Interestingly enough, installing the difficult to understand tool (FlowViewer) is faster with fewer difficult bits to put together.
Some pointers of places that have lists of available netflow analysis (or other) tools.