Le 24 juin 2010 à 16:19, Phil Mayers a écrit :
> On 24/06/10 14:53, Frédéric LOUI wrote: >> In our lab setup, we have 2 full internet feeds (IPv4 / IPv6) thus ~ >> 670000 + v6 prefixes Netflow v9 is working without any issue so far >> (without sampling). However we did not test netflow v9 with MPLS >> features. > > Sampling on the PFC/DFC platforms probably doesn't do what you want it > to do. It still captures all flows; just only exports a subset. This > saves neither TCAM nor significant CPU at the export side. > >> According to my technical representative, on 8x10GE card with 3CXL, >> NDE is handled by the linecard itself (i.e de-centralized NDE instead >> of centralized NDE from the main RSP board) Not sure about " 4x10GE >> card upgraded with 3CXL" though > > Under most conditions, DFC-equipped linecards export the netflow themselves. > > There are some conditions where the netflow PDUs have to be post-processed by > the RP. Amongst others, if you want source/dest AS numbers in the flows, > these are added in software by the RP, eliminating the benefits of > distributed NDE. Thanks very much for the precisions. > >> >> I'd be happy to have some REAL figures regarding MPLS aware netflow >> with 760x/3CXL :-) > > Figures about what? Sorry, what I meant by *figures*: - How much (additional?) netflow traffic (bandwidth) can be transferred by NDE - How many (additional?) flows the BOX can handle ? - What is the impact in term of resource and linecard CPU when activating MPLS aware netflow ? Has someone experienced performance degradation on a 3CXL 760X when activating MPLS aware netflow ? Thnks in advance for your feedback :-) -- Frederic > > Also, be aware that DFC/PFC platforms (until recently) only generate flows > for packets which ingress as plain IPv4. MPLS LSPs don't generate netflow, > either in label-swap or label pop. > > This may mean you only see flows at the ingress PE. > > There is a feature in some releases (e.g. SXI4 for 6500/sup720, which is > basically the same underlying hardware) which claims to do "egress mpls > netflow", where flows are generated for final label-pop i.e. on egress PEs > too. It didn't seem to work when I tried it. > > This may all be very different if 7600 WAN cards are in the mix. > _______________________________________________ > cisco-nsp mailing list [email protected] > https://puck.nether.net/mailman/listinfo/cisco-nsp > archive at http://puck.nether.net/pipermail/cisco-nsp/ _______________________________________________ cisco-nsp mailing list [email protected] https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
