Hi Zenon,

On Tue, Oct 27, 2009 at 09:44:52PM +0200, Zenon Mousmoulas wrote:

> I think so, yes. It reminds me somewhat of RPF for IP multicast.
> 
> Just so I'm sure I understand it right: You lookup the source IP
> address or network (for prefix-aggregated flow records) in the BGP
> RIB and you then match the next-hop (among other options) against
> this map. Correct?

Correct. RPF checks the input interface. Which is a check supported
by the bgp_peer_src_as_map. Additionally, you can check the next-hop
(or alternatively as you said the first AS in the AS-PATH - although
this is not currently supported) in scenarios where you have multiple
peers off a single interface (which is the typical Internet Exchange
scenario). 

> I can't figure out how the lookup helps when there is asymmetric
> routing. It runs at our end and not the far end, so it can't reveal
> the actual entry point of the traffic. Right?

The BGP next-hop check on the source IP prefix helps by limiting
false positives in case of asymmetric routing - compared to getting
peer-AS information via NetFlow from the router (as the router only
does a FIB lookup). 

The ideal solution would be having MAC addresses exported via NetFlow;
NetFlow v9 is up to that; does your C12k allow to get such L2 info
via NetFlow?

> What about peer_dst_as? Not as complicated as peer_src_as, but is a
> RIB lookup used to calculate it?

Correct.

> >Broader question is good: correlation of NetFlow and BGP is done
> >by looking up source and destination IP prefixes in NetFlow into
> >the BGP RIB. If you aggregate IP addresses logically to the network
> >boundaries - then it's still OK.
> 
> What do you mean by the last sentence? For router-based aggregation
> the only scheme I can think of which still shows (summarized) IP
> address space is prefix-based aggregation.

That's what i was meaning. Prefix-based aggregation would work fine
in correlating NetFlow flows and BGP RIB data.

> OK I see. So AS-aggregated netflow is practically useless (for now)
> if you are going to use the BGP features.

Correct.

> >IHMO, it would be good to put some efforts in this direction -
> >specifically, before even developing something general-purpose, it
> >should be seen how much accuracy is affected by rather common
> >de-aggregation or prefix load-balancing practices on the internet:
> >ie. i have one AS, two /18 allocations, two upstream provider: i'll
> >advertise a /18 here and one there.
> 
> I'm not sure I understand this, perhaps you'd like to further elaborate.

If an AS spreads its IP address space across multiple AS-PATHs for
whatever purpose (load-balancing mainly) and NetFlow is AS-aggregating
at the router (so you don't have IP prefix visibility anymore), what
is the correct AS-PATH for a given aggregate? You can't say for sure
as there are multiple choices. 
Effectiveness (or precision) of AS-aggregated NetFlow depends on how
diffuse are practices like this (spreading or de-aggregating the own
IP address space) by service providers on the internet. 

> In any case, the motivation for router-based aggregation is that
> nfacctd will have to process less flow records, thereby increasing
> the scalability of the whole setup, compared to unaggregated
> netflow.

Fully see the point.

> Both. The router advertises itself as the next hop for any
> particular prefix it has not originated, instead of the original
> next hop. It is used to 'blind' your peer so that it doesn't need to
> also have a route installed for the actual next hop (which could be
> anywhere).

Setting the next-hop-self in case of eBGP peerings is diffuse practice,
expecially in MPLS networks, and it's OK (all next-hops are from your
own infrastructure; well, let me add a disclaimer: this is in theory).
Setting the next-hop-self to iBGP peers, yes, has a blinding effect and
hence its use should (must?) be avoided to the collector. 

Cheers,
Paolo


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

Reply via email to