[c-nsp] MLPPP jitter over T1 links with large latency difference

[Andy Ellsworth]

We have a MLPPP bundle built from six international PtP T1s that are
provisioned over various diverse paths, which results in a latency
differential ranging from ~90 ms for the fastest link to ~120 ms for the
slowest link.

When trying to push RTP traffic across the MLPPP bundle, the _lower_ bound
on jitter is roughly equivalent to the latency spread, which kinda makes
sense given the round-robin nature of MLPPP.

Juniper has solved this problem by providing an option to do hash-based link
selection for high-priority traffic like voice/video:

http://www.juniper.net/techpubs/en_US/junose10.0/information-products/topic-collections/command-reference-n-z/ppp-hash-link-selection.html

With hash-based, each RTP flow would stay on a single link, and would not be
subject to the jitter problems associated with per-packet round-robin
distribution. However, you still get per-packet for best-effort traffic,
which keeps the distribution mostly nice and even.

Anyone know of a way to achieve similar functionality in Cisco land? Or
perhaps another way to solve the problem?

If it matters, US side is a ASR 1004 w/SPA-8XCHT1 running 2.4.3, int'l side
is a 7206 VXR + NPE-400 w/PA-MC-8T1 running 12.4.

Thanks.

-Andy