[c-nsp] Ignoring 7200 Bandwidth Points

[Justin Shore]

I've got a 7206VXR w/ 4 PA-A3-OC3SMI PAs serving a couple thousand PVCs 
of RBE DSL.  I have another 2x OC3s on a 3660 doing the same thing only 
with less PVCs.  The 3660 crashed twice earlier this week in one day. 
Once was on its own.  The second was in the middle of a sh tech.  I sent 
the crashinfos to Cisco but didn't get much info back.  The 3660 is EoS 
as of 12/08.  There are problems on the 3660 and the 2 DSL systems that 
it served and the problems don't appear to be random.  Some DSL 
customers (and these persist over time and through reboots) aren't able 
to get an IP.  DHCP is on the 3660.  In the debugs I see the DISCOVER 
come in and the OFFER get sent back out.  Some time later (15-60s) I get 
another DISCOVER.  It's as if the CPE never received the OFFER, timed 
out and sent another DISCOVER.  The carrier equipment beyond the 3660 is 
AFC gear.  They've rebooted DSL cards, CPUS, LETs, OC3 cards, etc but 
nothing fixed it.  I rebooted the router in the end and the reports were 
that the problem went away.  However the customers are now calling back 
in reporting issues again.  And again I see OFFERs go out and never get 
a REQUEST back.  One-way communication?

I'm considering moving the OC3s over to the 7206 (NPE-G1).  I'll have to 
pick up some PA-A3-OC3SMI PAs to do it.  I believe my 4 existing OC3s 
put me at the max on bandwidth points.  However throughput on them is 
very low.  Average CPU on the G1 is less than 10%, throughput on the 
OC3s averages only 10-15Mbps with peaks maybe twice that.

I believe I read that I can ignore the bandwidth points warning and load 
up the chassis if the PAs are running at sub-rates.  Would 6x OC3 PAs 
running at the low rates I described be a problem for a G1?  I can pick 
up a couple extra OC3 PAs and have them here in a few days if that's the 
case.

Thanks
  Justin