First post here as we're at a complete loss on this one. I hope it's not a simple answer in a way ;o) We are bonding two LLU DSL lines using MLP. Our LLU provider supports MLP bonding and we have a few other customers working well on Cisco 1841s, although not such high sync speed lines as this problem site.
So the lines work well with no interface errors and sync speeds are very good and evenly matched between the two lines (approx 14Mbps downstream). It's all good - it's great in fact except that it just doesn't work properly! By that I mean we're not seeing the downstream throughput we'd expect. We actually get the downstream throughput of less than a single DSL line, so about 12Mbps. Upstream bonded throughput is fine and in line with the sync speeds.
Both circuits are 'active' in the PPP multilink bundle and I see 3 sessions on our core LNS Cisco 7301 (c7301-boot-mz.124-2.T.bin) - i.e. 2x circuits + 1x bundle. We've checked the circuits individually and 'actual' throughput (using NetPerf software) is similar for both lines and in line with the sync speeds.
We are seeing quite high CPU (50%) on the Cisco 1841 (c1841-ipbasek9-mz.124-24.T1.bin) at the customer premesis, but having tested a Cisco 2951 on the customer premesis with two new HWIC-1ADSL-M cards, this is not the cause. The 2951 ran at 5% CPU whilst we experienced the same problem.
We've checked the setup of both Cisco CPE and LNS with our LLU provider and they are happy with the MLP config. They themselves have been able to bond two similarly sync'd Annex-M DSL lines and get 25Mbps throughput on a 1841.
Cisco configs and outputs attached.
What Cisco wants is the business case. Show up with a big sale prospect and keep them on the hook to deliver whatever functionality.
If you don't have that, the only you can do is the "Product Enhancement Request", meaning they file and may never look at it.