Well I've done a few download tests and it seems the modem only sustains
about 3.9 kilobytes per second, so obviously the speed is being
downgraded even further after connect. Line quality probably having
some effect and I'm still dubious about the quality of the modem itself
- as I say, other modems connect at a higher initial speed to the same
ISP. Then again, I didn't check their subsequent download rates...
perhaps they likewise downgraded.
Anyway, I'm not going to concern myself further about it - like you say,
it's little difference. It's a reliable connection and 3.9 kps is
plenty - I still remember borrowing a relative's 28.8k modem when they
were the hottest new thing and it was like driving on the autobahn.
OK, a bit of positive follow-up - results have changed since my initial
postings. Things seem to be fairly variable. Right now I have a reported
connection string of 50666. I have written a small KDE Panel applet that
has a graph and queries the ppp interface for stats. I'm downloading using
gtk-gnutella with several concurrent sources so the connection is
saturated.
The current 15-minute average is 5722 bytes per second which is equivalent
to 45776 bits per second.
A short while ago, however, my 15-minute average was 8470 bytes per second,
equivalent to 67760 bits per second. I had 1 or 2 second bursts of a lot
more, including a maximum of 12850 bytes per second (102800 bits per
second, pretty close to the maximum 115200 of my UART port).
So my problem is obviously not that the modem can't operate at high rates,
and the comments by posters that initial connection string is somewhat
irrelevant are spot on - considering that with a very similar connection
string I've had subsequent sustained transfer rates that are so far apart
that one is more than twice the other (earlier in the thread I reported an
ftp download of 3.9 kilobytes per second - 31948 bits per second - and the
http://www.pcpitstop.com test returned 29 kilobits per second - 29696 bits
per second).
So to summarise:
Date Connect Test Speed (bits per second)
25/5/2005 50666 ftp download 31948
26/5/2005 49333 pcpitstop 29696
04/6/2005 50666 gnutella dnld 45776
04/6/2005 50666 gnutella dnld 67760
I won't focus on the spikes, but the average of 67760 bits per second is
well above the reported connection speed of 50666 and well above even the
theoretical maximum of 56000 bits per second. I'm assuming that
compression plays a major part in this.
I'm also not sure how stop bits and the like relate to reported rates. If
the modem reports that it is connected at 50666, does this mean that it
will transfer my user-data of x bytes in (x * 8) / 50666 seconds, or will
it add stop/start bits or other gunk so that the actual transfer time of my
x bytes becomes (x * (8 + y)) / 50666 seconds, where y is the number of
start/stop/other bits added, or some other similar formula? I'm ignoring
compression.
If the second is true, then connect rates don't really represent
user-transfer rates and are actually over-inflated, so assuming a single
stop bit, and that my second formula is accurate, a reported speed of 50666
would actually concur with an experienced speed of 8/9 * 50666, or 45036.
That could be another reason why connect strings do not match experienced
speeds.