NetFlow version 9: egress vs. ingress
Posted in NetFlow, NetFlow Analyzer, Network Traffic Analysis, Scrutinizer on June 4th, 2009 by mike@plixer.com
I’m doing some more work lately with Wireshark and Scrutinizer v7. I thought that the topic of egress vs. ingress might be interesting to some readers. NOTE: Egress is only available in Cisco NetFlow v9 and not NetFlow v5.
IPFIX or NetFlow v9?
In theory, ingress and egress should work the same in IPFIX, which is based on NetFlow v9, but they are certainly different. Although they are very similar, don’t let any company tell you they are exactly the same. Many collectors that work with NetFlow v9 will puke when they receive IPFIX. Scrutinizer handles both with ease. Nortel supports IPFIX, as does/did Avici, which is now Soapstone Networks, Inc. Other vendors, such as Adtran and Enterasys, support NetFlow v9.
One annoying area where IPFIX and NetFlow v9 differ is in the labeling of fields: NetFlow v9 has ‘IN_BYTES’ and IPFIX labels the same field ‘octetDeltaCount’. IPFIX probably renamed it because when talking about egress flows, IN_BYTES is sort of misleading.
Ingress vs. egress differences
NetFlow v9 Ingress is collected on traffic going into (i.e. inBound) an interface. This is how NetFlow v5 collects data. To figure out outBound traffic volume, ingress must be collected on all interfaces and the reporting software then displays outbound traffic. What goes in must go out, right? Ya, usually.
NetFlow v9 Egress is collected on traffic going out (i.e. outBound) of an interface. Generally, it is used in combination with Ingress, but it doesn’t have to be. I’ll dive into this a bit more.
Why collect with egress?
Why collect with egress, if ingress worked so well with NetFlow v5? Because hardware such as WAN optimizers compress data. Traffic compression with Cisco NetFlow means that what comes in 100 bytes might go out as 50 bytes. If only using ingress flows, the NetFlow reporting software will show 100 bytes outbound, even if it was compressed to 50 bytes. GASP!!! This is because it was calculated using ingress flows.
Tell me the truth!
If the router is exporting both ingress and egress and the NetFlow monitor can report on both without overstating utilization, you can see how much of each flow is being compressed. It’s pretty slick, but it requires that the NetFlow collector understand what is known as the flow “Direction”. If the field in the NetFlow v9 packet is a 0, then it is an ingress collected flow. If the field is a 1, then it is an egress collected flow.
Ingress Flow with IPv6 (the same with IPv4)
Egress Flow with IPv6 (the same with IPv4)
The network traffic reports produced by the NetFlow analyzer need to be intelligent when dealing with ingress and egress flows. I feel that dynamically figuring out flow direction in mixed NetFlow v9 ingress egress environments is crucial, especially if the customer has hundreds of routers. If you are just setting up ingress, I would keep this blog in mind: “ip route-cache flow or ip flow ingress… Which do I use?”
Something else to think about
NetFlow traffic analysis is going to be taken to another level as Flexible NetFlow matures. Perhaps we’ll see it take advantage of what NetFlow v9 calls ‘OUT_BYTES’. (IPFIX, needing to be different, calls this same field ‘postOctetDeltaCount’.)
Now you might ask: how is it related to ingress or egress? Stay tuned…
Michael PattersonScrutinizer Product Manager
Follow Me on Twitter
[...] **NOTE** With NetFlow v5, we only had the option to monitor inbound statistics using the ip flow ingress command. However, with the release of NetFlow v9, we now have the option to monitor traffic leaving each interface via ip flow egress. Check out this blog which tackles the question: Which one is better to use? Ingress or Egress? [...]
Hello,
Some of your posts are awesome, and I really enjoy reading them. I have a very basic question regarding ingress and egress neflow data. How does CACTI or MRTG plot the interface in and out using the netflow v5 only. May be you would have explained this earlier but it would be great if you could explain in more detail if possible with an example.
Thanks,
Venkatesh
Hello Venkatesh,
Cacti and MRTG use SNMP to query the ifInOctets and ifOutOctets of a device to display in and out traffic. I don’t think they use NetFlow at all. The database of these tools is not practical for NetFlow storage. Storage and aggregation are paramount issues to consider for NetFlow analysis and reporting. Here is how we do it:
http://www.plixer.com/blog/scrutinizer/unprecedented-netflow-and-sflow-archiving/
I hope this helps.
Mike
[...] flow, which is something that NetFlow v5 can’t do. You need to read this blog on “NetFlow version 9: Egress Vs. Ingress” to understand the value of Egress flows. What I want to stress in this blog is that [...]