Hi Paolo,

On Jun 20, 2013, at 7:23 PM, Paolo Lucente wrote:

> Hi Osama,
> 
> Very good to know, great stuff. About the extra fields your colleagues
> would be seeking for: it would make a lot of sense to include support
> for field type #90, do you have it populated in your NetFlow export? If
> yes, i would highly appreciate to receive privately a brief trace which
> includes the field (even if you kindly have some room to implement this
> yourself). Thing is, in IPFIX (*) this field is defined as octetArray
> which, in turn, smells variable-legth (which in turn means it requires
> IPFIX rather than NetFlow v9 to be implemented properly). Let me know,
> i'm very curious. I did not manage to see it populated in any of the
> egress NetFlow traces i've been looking through.

You are right. The #90 is not exported by our equipment (Cisco). It was merely 
a _wish_ rather than a limitation.

> 
> Curious also about the other set of fields you were speaking about: i
> saw them only included as part of some NSEL exports but always empty:
> their size, unsigned32, does not coincide with the size of a MPLS RD so
> i wonder what is being actually encapsulated in there.

We already export #234 and #235 capturing the ingress and egress VRF ID 
respectively (Cisco). You are right, they are 32 bit integers. I did link the 
results with SNMP (Cisco) OID 1.3.6.1.4.1.9.9.711.1.1.1.1.2 to get their names. 
Works well. I can provide you with some pcap dumps if you want to have a look.

Do you think it is generic enough of a use-case to be incorporated in pmacct?

> 
> Cheers,
> Paolo
> 
> (*) http://www.iana.org/assignments/ipfix/ipfix.xml
> 
> On Thu, Jun 20, 2013 at 03:09:09PM +0200, Osama Abuelsorour wrote:
>> Hi Paolo,
>> 
>> I've just tested the new CVS version. Indeed the three primitives 
>> mpls_label_bottom and mpls_label_top show in our aggregations. The 
>> combination of this and *_iface primitives would help us classify traffic 
>> from the core side.
>> 
>> I'll work a bit more on it to come up with a process that combines with it 
>> SNMP (for interface indices and labels resolution). If it is generic enough, 
>> I will contribute it.
>> 
>> p.s. Some guys in my team are getting a bit more greedy: They were wondering 
>> about the mplsVpnRouteDistinguisher (#90) and VRF stuff (#234, #235, #236). 
>> If I have time next week, I might just add them myself (unless you are 
>> already working on that).
>> 
>> Thanks again for the great work!
>> 
>> On Jun 9, 2013, at 12:49 AM, Paolo Lucente wrote:
>> 
>>> Briefly to follow-up on this, to say:
>>> 
>>> * support has now been introduced for MPLS_TOP_LABEL_ADDR (NetFlow v9/
>>> IPFIX field type #47) which is mapped to peer_dst_ip field, ie. where
>>> typically BGP next-hop is written to. It would be nice if you could
>>> confirm presence of this field in your export (maybe along vendor and
>>> model information). 
>>> 
>>> * support has also been introduced for top and bottom MPLS labels (top
>>> is intuitively MPLS_LABEL_1 whereas bottom is determined dinamically
>>> by checking the BoS bit): mpls_label_top and mpls_label_bottom. If a
>>> flow has only a single label the value of twos will coincide. Having
>>> visibility of the top label value is especially useful to resolve
>>> the BGP next-hop in cases in which MPLS_TOP_LABEL_ADDR field is not
>>> available; the mpls_label_bottom is key in resolving which L2/L3
>>> MPLS VPN, if any, traffic belongs to.
>>> 
>>> * To further help in distinguishing VPN traffic, a derived primitive
>>> has also been made available: mpls_stack_depth.
>>> 
>>> This is all available pulling the code from the CVS. Hope this helps,
>>> looking forward at receiving feedback.
>>> 
>>> Cheers,
>>> Paolo
>>> 
>> 


_______________________________________________
pmacct-discussion mailing list
http://www.pmacct.net/#mailinglists

Reply via email to