Re: [IPFIX] comments on draft-scholz-ipfix-rtp-msg-00

[Hendrik Scholz <[e-mail]>]

Hi Hadriel,

On 03/26/2012 10:29 PM, Hadriel Kaplan wrote:
> 1) From a IPFIX IANA allocation perspective, it makes sense to get IANA-assigned IE numbers for things that could be exported from different vendors, instead of using PENs.  Quite a few of the IEs in this draft are fairly vendor-specific, however.  I mean in the sense that few media-monitoring vendors report the same type of information as in this draft.  I think there is some commonality across vendors, but it's a smaller list.  How do you want to handle that?  Do we get in a room and agree on a common list?

I'd be happy to separate the information elements into the ones that
are relevant to (or can be generated by) a larger audience and the more 
specific ones. Brian also mentioned something in this direction.

> 2) Some of the IEs in here imply the IPFIX records could be generated per-RTP packet.  I don't think you want to do that.  Some media-monitoring systems handle multiple millions of RTP packets per second.

Per-RTP packet record generation is not intended but it must be
possible to generate flow records for sections (e.g. 10 seconds)
of an RTP stream. In case of transient problems a higher
granularity will show isolated issues (e.g. a short period
of packet loss).
The records should allow mediators (or collectors) to aggregate
data from multiple records into a longer RTP stream.
So as a worst-case an IPFIX recording describing a single RTP
packet is possible but should not be the default.
The same mechanism should also apply to the -sip-msg records.
It should be possible to describe the outline of a session
as well as push a single SIP message for troubleshooting applications.

> 3) Some of the IEs imply there's a single SSRC for a flow - are you assuming a flow is by definition the 6-tuple of IP src/dest, port src/dest, transport type, and SSRC?  In other words, if there are two or more SSRCs in the same 5-tuple, would they be separate IPFIX flows? (they're only one "RTP" session from an RTP and SDP perspective)

Yes, a separate SSRC in between two nodes should be recorded
in separate records. After all these streams would most likely
belong to separate calls.

Also note that draft-akhter-opsawg-perfmon-ipfix-02 is available
and I duplicated some of the work done by Aamer.

Cheers,
  Hendrik

-- 
VOIPFUTURE GmbH   Wendenstraße 4   20097 Hamburg   Germany
Phone [phone]    Fax [phone]
Email [e-mail]   Web http://www.voipfuture.com

CEO Jan Bastian
Commercial Court AG Hamburg   HRB 109896
VAT ID DE263738086