Re: [pim] PIM yang model counter 64/32 bit

[Stig Venaas <[e-mail]>]

Hi

To me it seems like a great idea to make the counters 64 bit. I expect
there will be pim implementations using 64 bit counters, and I don't
think backwards compatibility with 32 bit implementations is an issue
since the counters are read-only. If you do this for this draft,
please make a similar change in the other multicast YANG models
(igmp/mld/msdp/snooping).

Stig


On Wed, Mar 7, 2018 at 8:27 AM, Anish Peter <[e-mail]> wrote:
> Hello PIM group,
>  We are doing pim yang model and came across this question of whether to use
> a 64/32 bit counter for packet statistics. To the best of our knowledge all
> the vendor models support 32bit counters.
>  But given that we are having more scalable implementations, we had a
> feeling that a 32bit counter may rollover too fast. To give some examples.
>
> Right now now we have routers with more than 10K pim interfaces supported.
> In such case even if we consider them to be p2p always, then it would mean
> 10K pim nbrs. In this case even if an aggressive hello timer of 10 seconds
> is configured then it would transmit and receive about 60K pim hellos per
> minute. This leads to about 86.4million hello per day. This way a 32bit
> counter could roll over in less than 50 days. Half a decade back I had
> tested pim for 4K interfaces with 1s hello interval. With this the same
> counter may roll over in about 12 days time.
>
> Similarly a pim null register and register stop too for a scale of 15K group
> could easily roll over in less than 198 days.
>
>
> The vendor counters are 32 for historic reasons and may not directly get
> impacted by a stats model supporting 64bit.
>
>  Please share your opinion.
>
> Thanks,
> pim yang model authors
>
>