IPFIX session at IETF 68

Quite a lot of things that were discussed during the session are relevant to our work on FlowMon. Here is my account.

The session took place on 20 March 2007 (minutes).

In general, this working group and the related PSAMP group generates many outputs that are important for our development, and not only of FlowMon. We should thus follow these groups more closely and perhaps actively partiipate in order to (i) avoid reinventing the wheel and (ii) promote more effectively our agendas and priorities. On the tactical side: while FlowMon is primarily a metering device and as such not really bound to IPFIX, we would greatly benefit from having at least a barebones IPFIX implementation. It seems that most of the research and standardisation work in flow monitoring is already being done in the context of IPFIX (even sampling algorithms and configuration), there are interoperability events etc. My conclusion thus is that we should accelerate the IPFIX implementation and with it in hands it will be much easier to participate in this community and get useful feedback.

Brian Trammel presented the draft on bidirectional flows and claimed that some support for them would be useful to implement in metering devices. Something to look at...

From my perspective, the most interesting presentation was Configuration Data Model (by Gerhard Muenz) dealing with draft-muenz-ipfix-configuration-01.txt. This draft contains an XML schema (written in XSD and intended for NETCONF) that describes the configuration of the whole flow monitoring chain. So its scope is broader than that of our FlowMon configuration schema, yet the comparison is quite interesting:

• They have no data slots for read-only statistics such as cache occupancy, counters of packets/flows processed at each step etc.
• They do not support per collector filtering of flow records
• They have pointers that join the configurations of subsystems in the concrete flow monitoring chain. For example, a configuration block describing a metering device points to another block that describes an exporter which takes data from the metering device. This might be useful for our configuration, too, in order to be able to represent more complicated setups.

I will definitely contact Gerhard and discuss these issues with him. For further details, see the HISTORY project webpage. In particular, the VERMONT toolkit looks very promising. Perhaps it could save us some effort in implementing IPFIX?