[arin-ppml] A challenge to the assumption that a big DFZ is a problem

Scott Leibrand scottleibrand at gmail.com
Mon Dec 14 14:19:34 EST 2009


Your analysis does apply, to some extent, to the lower end of the router 
market.  However, there are a number of higher-end routers (including 
pretty much anything that's more than 2U high and has more than a 
handful of 10G ports) where the limitation is in the custom TCAM used by 
the FIB.  Moore's law definitely applies there, but it's an entirely 
different architecture than general-purpose CPUs.

That said, most newer hardware currently supports about a million IPv4 
routes (or 1/4 as many IPv6 routes).  For those of us who only have to 
carry the DFZ table, that means we'll be OK for at least a few more 
years, and by then Moore's law should result in better faster cheaper 
replacements to handle the growth.  There are, however, networks out 
there that also inject a whole bunch of internal routes into their 
table, and they're a lot closer to the edge.

My own personal opinion is that we'll be fine as long as prefix growth 
continues at about the same rate it has recently.  But if we make a 
drastic change that results in a dramatic increase in the growth rate, 
that could get expensive.


On 12/14/2009 11:07 AM, Ted Mittelstaedt wrote:
> One of the fundamental assumptions that we all seem to accept many
> times is the "Ballooning BGP table/DFZ"  Much policy discussion
> seems to be centered around the idea that if the DFZ gets bigger
> it's going to cost bazillions of dollars for every ISP to upgrade
> equipment, yadda yadda yadda.
> I have to ask, however, is this assumption really technically
> accurate?
> Today I can walk into the store and purchase a PC that has a CPU
> in it that runs at a clock speed of at least 10 times of
> most routers, and has at least 10 times the amount of ram, for
> a quarter of the cost of the annual service contract for most
> DFZ routers  (let alone the hardware cost)
> Now, I think anyone who studies router hardware would probably
> agree that the reason routers have historically been so underpowered
> is that router vendors use older, less expensive, and more tried,
> technology AND there isn't a NEED for faster tech.  Why put a super
> powerful CPU or ASIC in the router when the purchaser only cares if
> the router can route at wire speed - and the wires consists of a
> few DS3's and 10/100 ethernet ports?  Older, cheaper tech can do those
> speeds while not even breaking a sweat.
> I have to ask, if the PURCHASERS of new router hardware were to tell
> the router vendors that they aren't gonna bother buying a router
> that cannot handle a half-million table entries in the BGP table,
> that the router vendors might just possibly see a need for the
> faster silicon - and step up to the plate and supply it?  God
> knows they charge enough money for the OLD tech they are supplying
> now.  Anyone heard of Moore's Law?
> I just have a hard time believing that when I can walk into a
> pizza place and drop a credit card down for payment - which carries
> just one of a billion or so possible numbers out there  - and
> get an approval on it in less than a second at the register,
> that the silicon and software doesn't exist that could handle
> a DFZ that's an order of magnitude larger than what we have now.
> Just a random thought....
> Ted
> _______________________________________________
> You are receiving this message because you are subscribed to
> the ARIN Public Policy Mailing List (ARIN-PPML at arin.net).
> Unsubscribe or manage your mailing list subscription at:
> http://lists.arin.net/mailman/listinfo/arin-ppml
> Please contact info at arin.net if you experience any issues.

More information about the ARIN-PPML mailing list