[c-nsp] DS-3 -> OC-3 upgrade

[Justin M. Streiner]

On Tue, 17 May 2005, Gert Doering wrote:

> On Mon, May 16, 2005 at 08:51:54PM -0400, Justin M. Streiner wrote:
>> On Mon, 16 May 2005, Gert Doering wrote:
>>
>>> What's "relatively low pps" for you?  A NPE-G1 should be able to shove
>>> through a OC-3 worth of 64byte packets with a load below 50% - unless
>>> you have "evil" features enabled, IP accounting being the worst of them.
>>
>> I saw this router buckle under a ~50kpps ICMP flood on top of the normal
>> traffic.  It dropped out of the IBGP mesh and lost all OSPF adjacencies.
>> The interfaces were still up, do the router still had enough juice to
>> handle keepalives, but it was pretty much unresponsive to all admin
>> connections - console was barely usable.  This was traffic directed at a
>> customer behind the router, not to the router itself.
>
> Hmmm.  This is not so good news.  A NPE-300 with netflow export can handle
> that load (but it will bring it up to 100%), so the NPE-G1 should quite
> easily handle it.  Even with 80kpps (maxing out the sender) I couldn't
> get the NPE-300 to lose its BGP sessions.
>
> My tests have been run with 12.2(18)S7, so YMMV.

You bring up a good point.  We installed that G1 not long after Cisco 
first started shipping them, and the code choices that supported the G1 
and the features we needed (nothing exotic, standard service provider 
stuff) were extremely limited and bug-ridden.

Of course that was nowhere near as much fun as just *getting* a functional 
G1 in the first place.  The first one came up from Cisco's plant in Mexico 
and sat in a Fedex truck for two weeks waiting to be cleared by customs. 
We get it and a few days after installing it, the router started 
reloading.  The TAC digs into it and eventually tells me it's a hardware 
problem - the usual "you have a revision B0 board and you need a D0".
Two more RMA's later, they finally get it right :-)

jms