Hi Ed,

Although you might be running into other typical C7600 issues with NetFlow,
ie. inaccuracy introduced by NetFlow TCAM space exhaustion (especially if
your RSP720 is not XL series), I concur with Brent suggestion to first of
all try enabling sampling. Also i'd definitely recommend to switch to its
packet-based version (by default it's time-based) (*, see NetFlow sampling
section). The pointed document should clarify how packet-based sampling is
working; also, by having collected myself NetFlow from C7600 boxes at some
ISPs, i can confirm you need to renormalize counters (factor in the sampling
rate: "nfacctd_renormalize: true" config directive) in order to get actual
values - so original flow bytes/packets counters must be getting manipulated
somewhere before/upon NetFlow export at the router. 

Speaking of pmacct: --enable-64bit is enabled by default since 0.14.3 as,
as you were correctly pointing out, it makes sense to.

Cheers,
Paolo

(*) 
http://www.cisco.com/en/US/docs/switches/lan/catalyst6500/ios/12.2SX/configuration/guide/nde.html#wp1164308

On Fri, May 24, 2013 at 04:39:58PM -0500, Edward Henigin wrote:
> Hi Brent,
> 
> >From what I've read on the SUP720 and RSP720, they do not support netflow
> sampling for creation/update, they only support "sampling" on export. I
> think the way that works is they randomly skip exporting some of the
> records. So I don't think that would actually impact the overflow issue...
> but I guess I should double-check that as an option...
> 
> Thanks for the idea,
> 
> Ed
> 
> 
> On Fri, May 24, 2013 at 3:47 PM, Brent Van Dussen <[email protected]> wrote:
> 
> >  Hi Ed,****
> >
> > ** **
> >
> > Does your RSP720 support sampled netflow by chance?****
> >
> > ** **
> >
> > -Brent****
> >
> > ** **
> >
> > *From:* pmacct-discussion [mailto:[email protected]] *On
> > Behalf Of *Edward Henigin
> > *Sent:* Friday, May 24, 2013 1:40 PM
> > *To:* [email protected]
> > *Subject:* [pmacct-discussion] Tips on dealing with overflowing 32-bit
> > fields?****
> >
> > ** **
> >
> > Hi y'all,****
> >
> > ** **
> >
> > I'm hoping that someone has some experience that might help.****
> >
> > ** **
> >
> > I'm using nfacctd to collect flows from a Cisco RSP720. After banging my
> > head against the keyboard for a few days, I realized I should have
> > configured pmacct with --enable-64bit. After re-building with that,
> > accuracy is dramatically improved, but I'm still finding bytes being
> > under-reported in numerous intervals.****
> >
> > ** **
> >
> > I believe the problem I'm running into is that the RSP720 is collecting
> > its data in a 32-bit field and that field is wrapping -- or, the netflow v5
> > packet uses 32 bits for bytes, and it's wrapping on export. In any case, I
> > think the byte count is lost before it leaves the Cisco.****
> >
> > ** **
> >
> > I'm using the "interface-destination" flow mask. I've tried using
> > "interface-destination-source" but that causes some CPU load on the Cisco
> > and flow creation failures.****
> >
> > ** **
> >
> > To have the highest resolution and to minimize the risk of netflow
> > creation failures, all the timers are set to the lowest:****
> >
> > ** **
> >
> >              enable timeout  packet threshold****
> >
> >              ------ -------  ----------------****
> >
> > normal aging true       32         N/A****
> >
> > fast aging   true       32         7****
> >
> > long aging   true       64         N/A****
> >
> > ** **
> >
> > 2**32 bytes in 64 seconds is only 537 Mbps. Doesn't work very well for
> > multi-gigabit traffic servers.****
> >
> > ** **
> >
> > Does anyone have any ideas on how to reduce or eliminate counter wrap on
> > the Cisco side for the bytes counter?****
> >
> > ** **
> >
> > Many thanks,****
> >
> > ** **
> >
> > Ed****
> >
> > _______________________________________________
> > pmacct-discussion mailing list
> > http://www.pmacct.net/#mailinglists
> >

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


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

Reply via email to