[j-nsp] recommended Netflow sampling rates?

[Ignacio Vazquez]

Hi Brian, Justin,
 we were doing some calculation about the error you could have trying to
protect at the same time the RE (not having more than 1 Kpps to the RE). I
wanted to have less that 5 % error because the sampling, and I realize this
could be obtain with more flexible sampling.

I know it could sound too conservative, but I think it is enough for us. My
result could be sumarized with this table:

Total traffic incoming          |   sampling recommended
x < 300 Mbps                    |     200
300 Mbps < x < 3 Gbps     |     2.000
3 Gbps    < x < 30 Gbps     |    20.000
30 Gbps  < x < 300 Gbps   |    200.000

Regards,

 Ignacio.





2008/8/31 Alexander Tarkhov <karabass at gmail.com>

> Hi Brian, Justin!
>
> This sampling story can be implemented in two different ways on M/T-series.
> One way of configuring the sampling output stanza is to utilize RE
> resources for exporting flows. That is the default (and free) approach
> when one of the daemons on RE - sampled - is doing the job of
> aggregating flows and sending them out via UDP.
>
> The other way of doing this is using services PIC (or some variant of
> it) with JFlow license installed for this purpose.
>
> The limitations of these two options are quite different, the intended
> purpose of the two approaches is also different. That might be a
> reason why there is no consensus here. You are probably asking about
> RE-based sampling on M/T-series or MX. Please clarify if this is
> correct, so that we don't mix these things.
>
> -Alex
>
> On Fri, Aug 29, 2008 at 6:55 AM, Brian Spade <bitkraft at gmail.com> wrote:
> > I'd be interested to learn this as well.  Also, how can you monitor the
> > router properly to know whether sampling is affecting performance
> (counters,
> > ram, cpu, etc.)
> >
> > /b
> >
> _______________________________________________
> juniper-nsp mailing list juniper-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/juniper-nsp
>