[arin-ppml] debunking the myth that Moore's law helps

[Michel Py]

> Matthew Petach wrote:
> You can get old 7k-class Cisco routers for $500 on eBay;
> why aren't more people just running large clusters of $500
> routers, then?  See the above cross-linking problem, and
> look at the rack space cost to put them in; at your typical
> facility like Equinix, at $1500/rack/month, if you can only
> fit 3 7k-class routers per rack, if the router only costs
> $500, you end up spending as much on space for the router as
> your hardware cost *every month*.

+1

You forgot the cost of electricity; colos charge or will charge for it.
These days, if you're not "green", you look "bad".

BTW, I have a mighty 7507 with dual RSPs to give away if someone cares
to drive and pick it up. 200lbs. 12U. $100 a month of electricity just
to power it up (not in winter, great heating system; in California
summer it's another story).


> Leo Bicknell wrote:
> However, with high end routers there are actually a number of
> new challenges.  Much like PC's, they have hit the "Megahertz
> Wall" and thus are moving to parallel processing solutions.

Indeed; which is why I started this thread in the first place. Where is
some kind of a sound barrier here; it does not mean it's a wall in the
sky that will rip your ears off if you try to break it, but there are
cost issues. If instead of DRAM we have to move to something equivalent
to CPU L1 cache today, I will remind everyone that the current size is
8MBytes, which does not go far on a DFZ router.


> The good news is packet processing is often easy to parallelize,
> the bad news is that it is all new code and hardware for the
> vendors.  Someone is going to have to pay for that development.

Guess who :-D


> I don't think these physics challenges are going to stop 100GE or
> other technologies, but they are going to disrupt the curve.
> We've gotten used to 10M ethernet replaced 5 years later by 100M
> for the same cost, replaced 5 years later by 1000M for the same
> cost, replaced 5 years later by 10000M for the same cost. It
> appears rather than seeing 100000M 5 years later and for the same
> cost, we're going to see it 7 years later and for twice the cost.
> (Ok, those are very crude estimates, but you get my drift.)

I do; I steer clear of 100G predictions.


>> Michel Py wrote:
>> Michel's law 0.9 says: "A device that is twice as fast or has
>> twice as much memory will consume twice as much bandwidth".

> Leo Bicknell wrote:
> http://singularity.com/images/charts/SuperComputers.jpg
> [SNIP]
> So no, twice the speed does not equal twice the bandwidth consumed.

[Love the 2025 requirement, BTW. Where do I sign up?]

Ok, come up with a formula? And read below, too.


>> Michel's law 0.9 says: "A device that is twice as fast or has
>> twice as much memory will consume twice as much bandwidth".

> Scott Leibrand wrote:
> I'm showing 14400 * 2^15 = 471859200, which is ~450
> Megabits, not Gigabits.  (Is that right?)

Same here.

> The doubling time may not be exactly 1.2 years (I've heard 18
> months), but it definitely looks like a similar curve to me...

Similar is the word. And although debating the actual value of the
"multiplying factor" sounds interesting, I would welcome more feedback
on the raw assertion that there is indeed a direct relationship between
Moore's law and Internet bandwidth.

Michel.