Re: Winmodem support, some performance tradeoff estimates

Adam J. Richter (adam@yggdrasil.com)
Sun, 16 Aug 1998 13:54:35 -0700


Executive summary: I was very wrong in my estimation of the
CPU power that a WebTV has available to run its soft modem in the
background. I apologize for the misinformation. Nevertheless, I
think winmodems would be economically optimal were there Linux support
for them now, and that this balance will probably shift more in favor
of winmodems in the future.

I wrote:
> The latest revision of WebTV runs on a 56k soft modem in the
>background on an 8MHz MIPS R4000 processor.

...and then argued that this meant that the CPU performance
cost of a soft modem was about $6.

Well, I was really wrong about this, and should have double
checked before posting. Thanks to Alan Cox for questioning this
number.

The WebTVPlus, which has the 56kbps soft modem as a background
task, actually uses a 167MHz R4640. So, my approximate upper bound on
CPU cost of a soft modem at $.75/Mhz by my previous argument would
then be $125.25 significantly more than the cost of a hardware modem
(correcting my previous estimatation of 8MHz x $.75/MHz = $6
estimate).

It might be worth finding out what is the slowest CPU on
which 56kbps winmodems currently run, as a better rough upper bound.
In the meantime, however, Alan has kindly included another basis for a
reasonable estimate of the CPU cost of a soft modem:

>Digitals own figures for a 200Mhz strongarm SA110 and their commercial
>software mode is 25% of the resources of the chip.

If 1 strongarm MHz = 1 Pentium MHz = $.75, then this would be
about $37.50 of CPU cost, or only a little savings from the cost
premium of a hardware modem, or maybe negative savings if the strongarm
is significantly faster per megahertz than the pentium.

On the other hand, the Cyrix cost per Pentium MHz appeared to
be something like $.25, which could bring that soft modem CPU cost
down to $12.50 ($37.50/3), and I think CPU price/performance ratios
will improve faster than those of hard modems, and you can use the
soft modem CPU power for other things when your modem is off, and I
think you CPU will likely continue in some active use for longer than
your modem. Also, if your usage pattern is that your CPU is basically
idle when you are downloading, your CPU cost is really only the CPU
cost of maintaining the software modem in an idle but connected state.

Anyhow, I will have to revise my claim that the CPU costs of a
winmodem are almost always vastly less than the price premium of a
hardware modem. It looks like it really depends on what level of CPU
you are using. For high end systems, with higher $/MIPS ratios,
WinModems may not yet be economically optimal. For most other systems
(Celeron, Cyrix), they might save you money already if there were
Linux support.

It also appears that the estimated 7.25 million Linux users of
the world are not going to collectively waste over US$100 million on
hardware modems, but I do think that that figure is probably still in
the tens of millions of US dollars as those that would save money with
winmodems find themselves unable to do so.

-- 
Adam J. Richter     __     ______________   4880 Stevens Creek Blvd, Suite 205
adam@yggdrasil.com     \ /                  San Jose, California 95129-1034
+1 408 261-6630         | g g d r a s i l   United States of America
fax +1 408 261-6631      "Free Software For The Rest Of Us."

- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.altern.org/andrebalsa/doc/lkml-faq.html