[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[ih] IPv4 address size debate

Well said.  Engineering is *always* about getting the job done with
minimal use of resources, and you always identify and focus on the most
expensive resources to drive the design.  The numbers have changed over
the decades, but the fundamental task remains the same.  

In addition to processor speed and circuit capacity, the other major
variable is cost.  That 1MHz computer back in 1980 cost a million plus
dollars - 1980 dollars.  The 3 GHZ machine on my desk today cost me $200
a few months ago.  For $1M I could get a whole bunch of them!  I think
the challenge now is how to use *lots* of cheap processors (and memory)
to do what you need to do.

When I was involved in IMP/X.25 networking, it was always the case that
the cost of the hardware (which we were selling) was insignificant
compared to the annual cost of the circuits.  It was common to do a cost
analysis which showed that the cost of a network (the boxes) would be
recovered in a year or so by savings on the communications lines.  So
the ongoing operational costs are just as important in the design
process as the costs of the equipment.  It all goes into the design

It would be interesting to see someone do an analysis of the changes
over time in computing and communication capacity, with cost (upfront
and recurring) and inflation factored in.  Or maybe it's been done.
Which is relatively cheaper now and back then - computing or

The goal is a moving target too.

We (at least I) never envisioned carrying lots of (or even any) video
data on the Internet, and certainly not high-def TV with independent
channels (no broadcast) to millions of end users.  

Computers back then had maybe a few dozen simultaneous users, and maybe
there might be hundreds of computers on the net.   It's amazing that
this stuff still works with millions... 

OTOH, I would never have predicted in 1980 that in 2009 there would be
millions of people typing one-line messages with their thumbs on tiny
keyboards with miniscule screens.   And on a device which is actually
just as capable of providing instantaneous real-time voice
communications!  What are they thinking...?

The numbers change, the user expectations change, the underlying facts
of life change.  Engineering design has to sort through all that and
come up with a good approach that will work now and for at least a
little while in the future.  

If it's too easy to get a design to work, then your solution is probably
overengineered and therefore wasteful.  But if the parts are so cheap,
you may not care.  The process of design becomes the larger cost, and
therefore that's what you minimize...

/Jack Haverty
Point Arena, CA

On Fri, 2009-11-13 at 20:26 -0600, Guy Almes wrote:
> Jack et al.,
>    I'm very much enjoying this thread.
>    I have a question about one part of this.
> Jack Haverty wrote:
> > Since I was one of the people arguing for fixed-length addresses back in
> > those days, I guess maybe it's time to explain why...
> > 
> > It all started with the network hardware, i.e., the I/O box that sat
> > between your computer and the network wire (Ethernet, IMP, whatever).
> > Your computer, whatever it was, was very slow and very expensive.  ...
>    You say that the early gateways were very slow and very expensive.
>    This is true, of course.
>    But think about how this has changed between 1980 or so and the present.
>    Computers have gotten faster, perhaps by something like 10,000 or so.
>    But if you consider that a "high-end serial pipe" in the 1980-85 era 
> was about 50 Kb/s and that such a pipe is 10 Gb/s, that's a speed-up of 
> 200,000 (math right?).
>    Even adjusting for the hardware difficulties of 1980, the modern 
> 10-Gb/s interface cards are difficult to build.  Oh, and expensive.  And 
> also, with DWDM, you wish you had a whole bunch of them.
>    Now I'm certainly grateful for these high-speed circuits.  The point 
> is not to complain.
>    But, to the degree that the IPv4 design tried to make things easy for 
> the limited processing hardware, that may still be the right thing.
>    (Now just what one means by "hardware friendly" is not quite the same 
> now as it seemed then, but I'll stick with the broader point.  The 
> capacity of the fibers and the demand for moving bits over them have 
> both increased faster than the processor speeds have increased.)
> 	-- Guy