[c-nsp] How does multicast multipath next-hop-based hashingactually work?

[John Neiberger]

On Wed, Oct 27, 2010 at 5:23 AM, Arie Vayner (avayner)
<avayner at cisco.com> wrote:
> John,
>
> In general the hash is done based on the source address (called S-Hash)
> There are enhancements to be able to base the hash on S and G (called
> S-G-Hash) or also taking the next hop into consideration...
>
> Take a look here:
> http://www.cisco.com/en/US/docs/ios/12_4t/ip_mcast/configuration/guide/m
> ctlsplt.html
>
> Some platforms still have it on the roadmap, so could be not yet
> available...
>
> Arie
>

Arie,

We're currently using the next-hop-based hash and still not seeing the
sort of balance we would like. Cisco TAC even had the developers take
a look at what we're seeing and were told that the behavior we're
seeing is expected, but we still don't know why. All we here is that
the hash is proprietary. We'd rather not leave the load sharing up to
guesswork and chance because we are running into situations where, for
example, one link in a four-link bundle will get a bunch of groups
assigned to it while another link might only get two groups. That
makes no sense, but TAC is saying that it is working as expected.

We're working with our dedicated Cisco engineer and Cisco Advanced
Services to get a better explanation. We really need to be able to
have more control of this traffic. This is a highly visible video
application and we can't have any packets dropped or lots of people
notice.

Thanks,
John