[arin-ppml] Policy Proposal: A Modest Proposal for an Alternate IPv6 Allocation Process

[Ted Mittelstaedt]

Seth Mattinen wrote:
> Ted Mittelstaedt wrote:
>> I have a comment/question on this..
>>
>> It appears the central rationale for this policy assumes that most
>> people are going to want to filter incoming BGP announcements,
>> presumably because the BGP table is going to grow rapidly and they will
>> otherwise run out of ram in their routers.  Is this assumption realistic?
>>
>> VISA and MasterCard corporation have devised systems that can handle
>> hundreds of millions of non-contiguous credit card numbers coming in for
>> approvals, from every corner of the globe.  I therefore have an
>> extremely difficult time believing that it is impossible to build a
>> router that cannot handle, say, 10 or 20 million BGP routes.  I also
>> have a difficult time believing that this cannot be done for the $50K to
>> $100K that Cisco and Juniper seem to think they can charge for a
>> fully-optioned BGP router.
>>
>> Today I can walk into the discount store and by a brand new PC with 2GB
>> of ram for under $350.  Yet, Cisco and Juniper are still including as
>> standard ram amounts, miserable, paltry amounts far smaller than that.
>>
> 
> Two points:
> 
> 1) On software platforms you're right. Last week I wanted to buy a 512
> MB upgrade for a 2811. It uses DDR 266 unbuffered ECC; old stuff. Cisco
> wanted $2,400 for it. They balked at me not wanting to buy their
> official RAM, claiming warranty voiding, etc. I said there's no way in
> hell I'm paying that much for old RAM so I bought some cheap-ass
> Kingston for $26 that works just as well.
> 
> 2) On hardware platforms TCAM is not DRAM.
> 

That is purely due to volume

Years ago, 8MB dram units were common, 512MB units were rare and
very very expensive.  Then software bloated and the increased
volume dropped the price of the ram.

Obviously, both Cisco and Juniper want to keep unit prices high
since they make better margins that way.  But, if the BGP table
did bloat to tremendous levels and both those companies continued to
select esoteric and expensive parts to use so as to keep their unit
costs high, they would just be opening the door to the Linux-based
companies (like Imagestream) who are waiting in the wings.  Cisco
is well aware of this, it is after all no surprise that they recently
reorganized their Linksys purchase and pulled all the high-end
linux-based routers (ie: rv082, rvs1000, etc.) out of that brand.
They wanted to quash that trend quickly.

If TCAM remains a niche product, the router companies would abandon it
if they were required to produce routers with many many gigabytes of
ram, and super-fast processors.

Remember the winmodem/hardware modem war?  Remember the 
software/hardware AC97 soundcard chip war?  In both of those wars
the supposition was that the winmodem and the software sound chips
were inferior to the hardware devices since the host CPU was loaded
down too much.  Both those wars were lost in favor of the software
parts because of host CPU speed advances.

20 years from today, we will have hardware so fast that everything
implemented in a hardware asic on a router today will be easily
possible to duplicate in software.

Ted