V.44 vs V.42 Compression
(Updated 17-Oct-03)

Can V.44 compression improve your dial-up Internet access?

This is a 46.6k V.92/V.44 connection with over 200k bps peak throughput:

This is a 50.6k V.90/V.42 connection with 145k bps peak throughput:

Wow. I've seen that it can - depending upon what you're transferring. 

What was I doing? The Net.Medic throughput displays above represent a 30-second snapshot of 2 connections to the same ISP with different local access numbers:

The 46.6k connection is to the new Earthlink (Level3) V.92 access number for Hilo, HI.
The 50.6k connection is to the soon to be discontinued Earthlink V.90 (UUNet) access number for Hilo, HI.

I used the Modemsite Softmodem in a Dell 4550 2.4 ghz Pentium 4 running XP to place the calls. 

The new number is a "virtual" local number - the server modem is in a remote location (Oahu), a CLEC (competivie local exchange carrier) is involved, and it's not done with "56k-modem friendly" technology, so the connect speed is lower - 4.0kbps lower than I get to UUNet's "real" local number. UUNet actually has server modems in Hilo - a "real" local call.

In both cases, I used Microsoft Front Page to publish changes to Modemsite on my server in California. This process involves transferring (highly compressible) data from the California server to my Hawaii PC. The instantaneous throughput (1/2-second) is 59.1kbps higher with the slower V.44 connection! (I captured close to the peak instantaneous throughputs on both connections. In both cases, the instantaneous throughput fluctuated 'normally' with the V.44 connection always significantly higher than the V.42.)

Now, this impressive difference won't always be in effect. And, Earthlink's new local number will actually be worse for compressed downloads - but not because of it being V.92. It will give lower connect speeds because of the impairments introduced by the telephone companies. 

Also see - Pavel Mitronov's page with detailed discussion and tests of various modems. (Note - the Mitronov tests were made with early V.92/V.44 client and Cisco Mica server modems.)

If your V.92/V.44 modem connects via a 'real' serial port you're in trouble:  most serial ports on PCs aren't able to handle real data at these rates. The 16550 UART typically found for COM ports is limited to 115.2kbps - and 20% of that rate isn't data - so, only 92,160bps of data can be achieved on a COM port set at 115.2kbps. (This can also be a factor with V.42 compression which, as you can see above, can also produce the potential for >115.2kbps transmission.) Even a 230.4kbps COM port wouldn't accommodate the throughput I got on my Softmodem's V.44 46.6kbps connect! Well-designed controllerless and softmodems aren't only much less expensive than hardware-controller modems, they can give better performance!