[c-nsp] ASR9k Bundle QoS in 6.0.1

[Robert Williams]

>> Bundles are tricky. Have you considered separate IP links instead? Cause then you could do MPLS-TE/SR tunnel over each link and forcing PWs into tunnels to get proper PW-to-Link distribution.

> For the core, this is what we're doing now.
> We've stopped running 10Gbps-based LAG's, and run multiple 10Gbps IP/MPLS links with equal IS-IS metrics. Makes the problem more manageable until we can move those links to native 100Gbps.

Agreed yes, we use this where there is limited capacity (outside our core) and we can't just drop another wavelength on it. In conjunction with RSVP selecting the appropriate path out of the x*1G or x*10G adjacencies. It works well although in our case, although it took a few adjustments to the auto-bandwidth parameters to stop path-swaps churning up entries in the logs as different customers become busy at different times.

Haven't yet deployed this to our core adjacencies because we are about 70/30 IP Transit Vs. MPLS+Ethernet services so the bundles work perfectly for our 'primary' traffic, this may well change in the future though and separate IP adjacencies per physical port is definitely the way to go until such time as the hardware can go a little deeper into the frame to hash.

Cheers guys!


Robert Williams
Custodian Data Centre
Email: Robert at CustodianDC.com
http://www.CustodianDC.com