Re: [tcpm] SYN/ACK Payloads, draft 01

On Wed, Aug 13, 2008 at 5:51 PM, Joe Touch <touch@isi.edu> wrote:
> I think you're quoting >40% by assuming that anyone who doesn't use
> broadband uses a modem (i.e., 100-57% = 43%), which isn't what those
> charts indicate. The direct metric from citation [1] is Fig 2, which
> shows less than 12% of homes using dialup right now, a figure that has
> been steadily falling for the entire timeline shown.

I believe that the number of homes with broadband is, indeed, 57%.
(Although it's a fair point that some might be using's others
unsecured wireless connections). On another reading of the report, it
looks like this is the number of homes were broadband is /available/.
There are /two/ figures labeled "Figure 2", so I'll refer to them as
the first and second:

I think the first fig 2 includes the numbers of people who are using
broadband connections at work and the second figure 2 has a much
higher number because it counts active users. Homes with broadband are
more likely to use it, it seems.

But, you're right, I should be using the number of active users rather
than the number of homes with broadband. I knocked up a model for the
losses experienced by an example website based on an extra round trip
time. I'm still trying to get permission to release it.

> [2] is a VERY old reference using a 10-year-old modem standard that
> lasted only two years (1998-1999). V.92 superceded it in 1999.

Yes, but I didn't find anything very good that was more recent -
people don't seem to study modems much anymore. Also, the V.92
standard increased the upload speed at the determent of the download
speed. It's not clear that V.92 would have have much, if any, effect
on latency.

> I measured RTTs from LA to Boston, and found 200-220ms. Pings locally
> were 140-150ms; the bulk of the increased ping delay was the modems.
> Unless you're going through a satellite or a very inefficient multihop
> radio system (e.g., anyone remember Metricom?), delays ought not exceed
> around 150ms + broadband delays, e.g., they ought to top out around 250ms.

It's about 75ms from SF to NYC, so that makes sense. However, 250ms
still makes a big difference. Like I said, I'm trying to get the
numbers released.

AGL

-- 
Adam Langley agl@imperialviolet.org http://www.imperialviolet.org
_______________________________________________
tcpm mailing list
tcpm@ietf.org
https://www.ietf.org/mailman/listinfo/tcpm