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

Leo Bicknell bicknell at ufp.org
Thu Dec 17 15:19:48 EST 2009

In a message written on Wed, Dec 16, 2009 at 11:41:59PM -0500, Scott Leibrand wrote:
> 14.4kbits * 2^15 is 450 Gigabits.
>    I'm showing 14400 * 2^15 = 471859200, which is ~450 Megabits, not
>    Gigabits.  (Is that right?)

Doh, quite right, 450 Megabits.

Still, you can't buy a GigE router at Best Buy, at all.  (As a minor
nit, you can buy a router with GigE, but not one that will route
even a fraction of it).

>    Gigabit Ethernet hardware today costs about the same (order of
>    magnitude) as what 14.4k modems cost in 1991.  You can even do GigE
>    over SMF (to achieve approximately the same distance range) for not
>    much more.  Once we get into the question of what medium is comparable
>    (twisted pair copper? cat5? MMF? SMF? PSTN? DWDM?), then obviously the
>    question gets a lot more complicated.  But as exponential doubling laws
>    go, I think this one holds fairly well so far.  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...

If the sole issue was the hardware interface; e.g. the cost of the
modulation hardware in the modem compared to the cost of a GigE chipset
you might be right; but that is not "bandwidth", that's simple hardware,
and back to Mores law.

However, the interface cost is only a small player in "bandwidth" costs.
Routing it takes more than just interfaces....


A 2 port 10 Megabit ethernet router was $5,495 list.  A Cisco 7301
router today (3xGigE) is ~$20,000 list.  That $5495 adjusted for
inflation would be $8331.  So in fact a router today costs "twice as
much" adjusted for inflation.

Of course, that's still not the entire picture.  Bandwidth costs
are "cumulative".  If your traceroute to say www.arin.net goes
through 7 routers, then the cost to "upgrade your connection" from
100M to 1000M for instance is 7 times the cost of upgrading an
individual box.  So you're having to pay your service provider for
6 of those (figuring you own the end box).  As the network footprint
expands, the number of routers in the path expands, and the cost
to provide you service increases as a result.

But wait, there's more.  In 1991 probably 99% of residental users
were wired for 14.4 modems (e.g. had a phone line).  You buy the
hardware and you're good to go.

In 2009, what percentage of homes can get GigE service?  1%?  I
doubt it rises even that high.  The fastest service I've seen for
residential users in the US is 100M FIOS, where available.  The
problem is that the way we've cost controled Ethernet chips down
to $1 parts is by requiring more from the cabling.  Rather than
using the copper in the ground and in the walls, it's 4 pair Cat
5e, or fiber.

Even fiber has not proven to be future proof.  Some of the original
fiber put in the ground in the early 1990's does not support 10G
WDM systems (at least, over any length) and had to be overlayed in
the 2000's with better quality fiber to keep up with network demand.

The question becomes, "what is bandwidth":

Chipset availability for higher speeds:  Follows Moores Law.
Router availability for higher speeds:   Follows Moores Law.
Router cost for higher speeds:           Expands with network diameter.
Availability of bandiwdth:               Expands with capital cable investment.

On the last point:


In 2007 the average broadband speed in the US was 1.9 Megabits.  If we
use the 14.4k modem in 1991 as our comparison again that's approximately
7 doublings, where as Moores law has taken us through 15 in the same
time.  If you're up on exponential curves that means the curves are
diverging, and at an ever accelerating rate.

       Leo Bicknell - bicknell at ufp.org - CCIE 3440
        PGP keys at http://www.ufp.org/~bicknell/
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 826 bytes
Desc: not available
URL: <https://lists.arin.net/pipermail/arin-ppml/attachments/20091217/7ebe3ec2/attachment.sig>

More information about the ARIN-PPML mailing list