serial port speeds >115k; PnP modem ID must contain modem substring

serial port speeds >115k; PnP modem ID must contain modem substring

[Tim]

I have a Diamond/Supra v90 ISA modem (model SUP2124) in my
Linux system running 2.4.20.  I've noticed that for the
type of file I commonly access (large html text table) 115k
port speed seems to be a significant bottleneck.  (These
files compress to <10% of raw size.  Transfer speed is
reported as ~10.5kByte/s, even if I reduce my modem speed
from the typical 45kbps to 28.8.)  The modem documentation
says it is capable of 230k port speed, so I tried specifying
230400 in my ppp config, which resulted in 9600 baud port
speed; stty returns an error.

At this point in my education, my guess is that the modem is
being treated like a 16550A without extended FIFOs or 230k
port speed.  If my modem was detected as a modem by Linux,
would setserial -a show something more?

# setserial -a /dev/ttyS3
/dev/ttyS3, Line 3, UART: 16550A, Port: 0x02e8, IRQ: 10
        Baud_base: 115200, close_delay: 50, divisor: 0
        closing_wait: 3000
        Flags: spd_normal skip_test

I am not a programmer, but have been looking at the comments
in serial.c:

There seems to be an ISA PnP device device search, the success
of which depends on finding the string "modem" in PnP id:
  dev->name or dev->bus->name must contain "modem" substring;

However, the modem in this system was installed with PnP
disabled and does not have modem in its string:
# ANSI string -->SupraExpress 56i Voice<--

What are the consequences of failing this test?  If the
string were changed, could the test succeed with the modem
jumper not in PnP mode?

There seems to be a "quickie test to see how big the FIFO is".
How can I cause the results of that test to be logged?

I now have CONFIG_SERIAL=y.
If I compile serial as a module, are there parameters that
can be added to modules.conf to enable available features
beyond what a 16550A has, or perhaps someone has patches
that would enable these features?

TIA,
Tim.

Re: serial port speeds >115k; PnP modem ID must contain modem substring

[whitnl73]

On Fri, 21 Feb 2003, Tim wrote:

> I have a Diamond/Supra v90 ISA modem (model SUP2124) in my
> Linux system running 2.4.20.  I've noticed that for the
> type of file I commonly access (large html text table) 115k
> port speed seems to be a significant bottleneck.  (These
> files compress to <10% of raw size.  Transfer speed is
> reported as ~10.5kByte/s, even if I reduce my modem speed
> from the typical 45kbps to 28.8.)  The modem documentation
> says it is capable of 230k port speed, so I tried specifying
> 230400 in my ppp config, which resulted in 9600 baud port
> speed; stty returns an error.

What error?  Quote exactly please.  If the doco says it can run
faster than 115200, it should also tell what you have to do to make it
do so.  (this may involve setting register(s) in the modem, and will
need in addition a setserial command to set baud_base (the serial
driver's understanding of how to set the divisor for am app (pppd,
FI) that requests a given speed) to get things to work
as you would expect they should).  I think you need to have programmed
one of these things by hand to understand the doco so that you don't
have to, but I don't know how to say it any better.
>
> At this point in my education, my guess is that the modem is
> being treated like a 16550A without extended FIFOs or 230k
> port speed.  If my modem was detected as a modem by Linux,
> would setserial -a show something more?

It was detected as a serial port.  Linux does not detect modems at
startup.  As far as the OS can see, a modem is just a serial port.
As far as the serial driver can see, it is a 16550a with a 16
byte FIFO.  The serial driver can detect more modern UART's such as the
16950, but yours seems to behave as if it were a 16550A.
You need something beyond what the serial driver does by default to get
the thing to run at a port speed >115200, I think.
>
> # setserial -a /dev/ttyS3
> /dev/ttyS3, Line 3, UART: 16550A, Port: 0x02e8, IRQ: 10
>         Baud_base: 115200, close_delay: 50, divisor: 0
>         closing_wait: 3000
>         Flags: spd_normal skip_test
>
...
>
> TIA,
> Tim.
>
Lawson
--
---oops---



________________________________________________________________
Sign Up for Juno Platinum Internet Access Today
Only $9.95 per month!
Visit www.juno.com

Re: serial port speeds >115k; PnP modem ID must contain modem substring

[David Lawyer]

On Fri, Feb 21, 2003 at 11:56:00PM -0500, whitnl73@juno.com wrote:
> On Fri, 21 Feb 2003, Tim wrote:
> 
> > I have a Diamond/Supra v90 ISA modem (model SUP2124) in my
> > Linux system running 2.4.20.  I've noticed that for the
> > type of file I commonly access (large html text table) 115k
> > port speed seems to be a significant bottleneck.  (These
> > files compress to <10% of raw size.  Transfer speed is
> > reported as ~10.5kByte/s, even if I reduce my modem speed
> > from the typical 45kbps to 28.8.)  The modem documentation
> > says it is capable of 230k port speed, so I tried specifying
> > 230400 in my ppp config, which resulted in 9600 baud port
> > speed; stty returns an error.
> 
> What error?  Quote exactly please.
It's clear that stty can't set 230.4k, likley because the 240.k feature
hasn't been enabled in the modem.
> If the doco says it can run faster than 115200, it should also tell
> what you have to do to make it do so.

Yes.  It will likely say (or imply) to use a driver they supply for
Windows only.  This will not be helpful for Linux.

> (this may involve setting register(s) in the modem, and will
> need in addition a setserial command to set baud_base (the serial
> driver's understanding of how to set the divisor for am app (pppd,
> FI) that requests a given speed) to get things to work
> as you would expect they should).

The above is not true in all cases since if you do the above, requesting
230.4k via stty etc., will simply send a divisor of value 1.  Some need
a certain high value of the divisor to work at 230.4k.

> I think you need to have programmed one of these things by hand to
> understand the doco so that you don't have to, but I don't know how to
> say it any better.

> >
> > At this point in my education, my guess is that the modem is
> > being treated like a 16550A without extended FIFOs or 230k
> > port speed.  If my modem was detected as a modem by Linux,
> > would setserial -a show something more?
> 
> It was detected as a serial port.  Linux does not detect modems at
> startup.

For ISA PNP (and PCI ?), the name of the device (the serial port built
into the modem) may include the name "modem" and Tim told me that
serial.c looks for the "modem" name for the ISA PNP case.

> As far as the OS can see, a modem is just a serial port.  As far as
> the serial driver can see, it is a 16550a with a 16 byte FIFO.  The
> serial driver can detect more modern UART's such as the 16950, but
> yours seems to behave as if it were a 16550A.  You need something
> beyond what the serial driver does by default to get the thing to run
> at a port speed >115200, I think.

The serial driver should ideally detect it by PnP, and enable it for
230.4k.  PnP identifies the device and then enables 230.4k for that
device.  But since Tim said in an email to me that had PnP disabled,
this can't possibly work.  
			David Lawyer

PnP Rage.

[Hal MacArgle]

Greetings: Does anyone care that I LONG for the good old days of
hardware with jumpers and silk screened instructions?

The "someone" who came up with PnP, eliminating ALL the problems we
had with port addresses and IRQ's has done a number on us judging
from the bazillion messages asking for help.. It seems to me the help
messages before paled to what we have now, not to mention all the
tools designed to help with the help.. Ugh.

Cheers,

    Hal - in Terra Alta, WV - Slackware GNU/Linux 8.0   (2.4.13)
                Proprietary  Formats  Unacceptable
.

Re: PnP Rage.

[David Lawyer]

On Sun, Feb 23, 2003 at 10:45:25AM -0500, Hal MacArgle wrote:
> Greetings: Does anyone care that I LONG for the good old days of
> hardware with jumpers and silk screened instructions?
> 
> The "someone" who came up with PnP, eliminating ALL the problems we
> had with port addresses and IRQ's has done a number on us judging
> from the bazillion messages asking for help.. It seems to me the help
> messages before paled to what we have now, not to mention all the
> tools designed to help with the help.. Ugh.

PnP enables the user not to worry about setting jumpers, etc. provided
of course that the various drivers are able to recognize the myriad of
PnP devices and PnP configure them.  For MS Windows the device
manufacturer provide such a driver.  But they usually don't provide such
drivers for Linux.  So what is a boon to MS Windows becomes a bane for
Linux.  For devices used primarily on servers such as several digital
modems on a card, Linux drivers will likely be provided since Linux has
about 25% of the server market.  But for devices primarily used on
desktops, the manufacturer is unlikely to support Linux and then someone
else needs to hopefully write the driver.  This is a major problem for
Linux.

			David Lawyer

Re: PnP Rage.

[Hal MacArgle]

Greetings and thanks for your input, but I neglected to mention, in
my message, that my PROBLEM's are with MSWindows setup and PnP.. I
have little problems with Linux once getting the hang of the 'isapnp'
conf routine.

The situation is moot now anyway as I've purged MSWindows from my life.
<grin> Thanks again,

Hal

On 02-23, David Lawyer wrote:
> On Sun, Feb 23, 2003 at 10:45:25AM -0500, Hal MacArgle wrote:
> > Greetings: Does anyone care that I LONG for the good old days of
> > hardware with jumpers and silk screened instructions?
> > 
> > The "someone" who came up with PnP, eliminating ALL the problems we
> > had with port addresses and IRQ's has done a number on us judging
> > from the bazillion messages asking for help.. It seems to me the help
> > messages before paled to what we have now, not to mention all the
> > tools designed to help with the help.. Ugh.
> 
> PnP enables the user not to worry about setting jumpers, etc. provided
> of course that the various drivers are able to recognize the myriad of
> PnP devices and PnP configure them.  For MS Windows the device
> manufacturer provide such a driver.  But they usually don't provide such
> drivers for Linux.  So what is a boon to MS Windows becomes a bane for
> Linux.  For devices used primarily on servers such as several digital
> modems on a card, Linux drivers will likely be provided since Linux has
> about 25% of the server market.  But for devices primarily used on
> desktops, the manufacturer is unlikely to support Linux and then someone
> else needs to hopefully write the driver.  This is a major problem for
> Linux.
> 
> 			David Lawyer
> -

Re: serial port speeds >115k; PnP modem ID must contain modem substring

[rich+ml]

Internal modems usually just have a simulated uart for software
compatibilty and baud rate is irrelevant (same for parity, stop bits,
FIFO, etc). Set to 5 baud and see if you get same performance.

== Rich

On Fri, 21 Feb 2003, Tim wrote:

> Date: Fri, 21 Feb 2003 15:09:39 -0500
> From: Tim <delt@gate-way.net>
> To: linux-serial@vger.kernel.org
> Subject: serial port speeds >115k; PnP modem ID must contain modem
>     substring
> 
> I have a Diamond/Supra v90 ISA modem (model SUP2124) in my
> Linux system running 2.4.20.  I've noticed that for the
> type of file I commonly access (large html text table) 115k
> port speed seems to be a significant bottleneck.  (These
> files compress to <10% of raw size.  Transfer speed is
> reported as ~10.5kByte/s, even if I reduce my modem speed
> from the typical 45kbps to 28.8.)  The modem documentation
> says it is capable of 230k port speed, so I tried specifying
> 230400 in my ppp config, which resulted in 9600 baud port
> speed; stty returns an error.
> 
> At this point in my education, my guess is that the modem is
> being treated like a 16550A without extended FIFOs or 230k
> port speed.  If my modem was detected as a modem by Linux,
> would setserial -a show something more?
> 
> # setserial -a /dev/ttyS3
> /dev/ttyS3, Line 3, UART: 16550A, Port: 0x02e8, IRQ: 10
>         Baud_base: 115200, close_delay: 50, divisor: 0
>         closing_wait: 3000
>         Flags: spd_normal skip_test
> 
> I am not a programmer, but have been looking at the comments
> in serial.c:
> 
> There seems to be an ISA PnP device device search, the success
> of which depends on finding the string "modem" in PnP id:
>   dev->name or dev->bus->name must contain "modem" substring;
> 
> However, the modem in this system was installed with PnP
> disabled and does not have modem in its string:
> # ANSI string -->SupraExpress 56i Voice<--
> 
> What are the consequences of failing this test?  If the
> string were changed, could the test succeed with the modem
> jumper not in PnP mode?
> 
> There seems to be a "quickie test to see how big the FIFO is".
> How can I cause the results of that test to be logged?
> 
> I now have CONFIG_SERIAL=y.
> If I compile serial as a module, are there parameters that
> can be added to modules.conf to enable available features
> beyond what a 16550A has, or perhaps someone has patches
> that would enable these features?
> 
> TIA,
> Tim.
> 
> -
> To unsubscribe from this list: send the line "unsubscribe linux-serial" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 

Re: serial port speeds >115k; PnP modem ID must contain modem substring

[David Lawyer]

On Sun, Feb 23, 2003 at 09:59:57PM -0800, rich+ml@lclogic.com wrote:
> Internal modems usually just have a simulated uart for software
> compatibility and baud rate is irrelevant (same for parity, stop bits,
> FIFO, etc). Set to 5 baud and see if you get same performance.

I set to 300 baud and couldn't connect.  Then at 1200 baud it connected but
download speeds were around 60 bytes/sec.  I estimated the received rate
from the /proc dir and it was close to 1200 baud.  So mine must have a
real UART.

Some modems brag that they can do 230.4k baud.  This is a serial port
speed.  So it seems like a real UART.  So what percent of internal
modems will operate at a high baud rate when one specifies low baud rate?

> 
> == Rich
> 
> On Fri, 21 Feb 2003, Tim wrote:
> 
> > Date: Fri, 21 Feb 2003 15:09:39 -0500
> > From: Tim <delt@gate-way.net>
> > To: linux-serial@vger.kernel.org
> > Subject: serial port speeds >115k; PnP modem ID must contain modem
> >     substring
> > 
> > I have a Diamond/Supra v90 ISA modem (model SUP2124) in my
> > Linux system running 2.4.20.  I've noticed that for the
> > type of file I commonly access (large html text table) 115k
> > port speed seems to be a significant bottleneck.  (These
> > files compress to <10% of raw size.  Transfer speed is
> > reported as ~10.5kByte/s, even if I reduce my modem speed
> > from the typical 45kbps to 28.8.)  The modem documentation
> > says it is capable of 230k port speed, so I tried specifying
> > 230400 in my ppp config, which resulted in 9600 baud port
> > speed; stty returns an error.
> > 
> > At this point in my education, my guess is that the modem is
> > being treated like a 16550A without extended FIFOs or 230k
> > port speed.  If my modem was detected as a modem by Linux,
> > would setserial -a show something more?
> > 
> > # setserial -a /dev/ttyS3
> > /dev/ttyS3, Line 3, UART: 16550A, Port: 0x02e8, IRQ: 10
> >         Baud_base: 115200, close_delay: 50, divisor: 0
> >         closing_wait: 3000
> >         Flags: spd_normal skip_test
> > 
> > I am not a programmer, but have been looking at the comments
> > in serial.c:
> > 
> > There seems to be an ISA PnP device device search, the success
> > of which depends on finding the string "modem" in PnP id:
> >   dev->name or dev->bus->name must contain "modem" substring;
> > 
> > However, the modem in this system was installed with PnP
> > disabled and does not have modem in its string:
> > # ANSI string -->SupraExpress 56i Voice<--
> > 
> > What are the consequences of failing this test?  If the
> > string were changed, could the test succeed with the modem
> > jumper not in PnP mode?
> > 
> > There seems to be a "quickie test to see how big the FIFO is".
> > How can I cause the results of that test to be logged?
> > 
> > I now have CONFIG_SERIAL=y.
> > If I compile serial as a module, are there parameters that
> > can be added to modules.conf to enable available features
> > beyond what a 16550A has, or perhaps someone has patches
> > that would enable these features?
> > 
> > TIA,
> > Tim.
> > 
> > -
			David Lawyer

Re: serial port speeds >115k; PnP modem ID must contain modem substring

[rich+ml]

On Sun, 23 Feb 2003, David Lawyer wrote:

> Date: Sun, 23 Feb 2003 23:11:36 -0800
> From: David Lawyer <dave@lafn.org>
> To: linux-serial@vger.kernel.org
> Subject: Re: serial port speeds >115k; PnP modem ID must contain modem
>     substring
> 
> On Sun, Feb 23, 2003 at 09:59:57PM -0800, rich+ml@lclogic.com wrote:
> > Internal modems usually just have a simulated uart for software
> > compatibility and baud rate is irrelevant (same for parity, stop bits,
> > FIFO, etc). Set to 5 baud and see if you get same performance.
 
> I set to 300 baud and couldn't connect.  Then at 1200 baud it connected but
> download speeds were around 60 bytes/sec.  I estimated the received rate
> from the /proc dir and it was close to 1200 baud.  So mine must have a
> real UART.
Or the emulation limits receive rda interrupt rate to the virtual baud and 
word size :) 

> 
> Some modems brag that they can do 230.4k baud.  This is a serial port
> speed.  So it seems like a real UART.  So what percent of internal
> modems will operate at a high baud rate when one specifies low baud rate?
Don't know, maybe none anymore, my experience with modem design is 
dated. 

Bear in mind that the baud rate is set as a baud clock divisor in the
16550, where the nominal baud clock is 115200Hz. So for example to get
9600 baud you program a divisor of 12. The divisor can't be less than 1,
so 115200 baud is the max.

But some uarts have double baud clock which doubles all baud rates.  
Setting the divisor to 12 gets you 19200 baud. The software won't
automatically know this, it will still be CALLED '9600'. (You can fix this
by setting setserial base_baud, but it only affects string-to-divisor
conversion, not the hardware).

So if you had a doubled baud clock then when you set '115200' you'd
actually get 230400. Except this doesn't seem to be the case. 

So either it's enabled via modem-dependent means (e.g. a jumper), or the
internal uart doesn't support it even though the core modem design does.

> 
> > 
> > == Rich
> > 
> > On Fri, 21 Feb 2003, Tim wrote:
> > 
> > > Date: Fri, 21 Feb 2003 15:09:39 -0500
> > > From: Tim <delt@gate-way.net>
> > > To: linux-serial@vger.kernel.org
> > > Subject: serial port speeds >115k; PnP modem ID must contain modem
> > >     substring
> > > 
> > > I have a Diamond/Supra v90 ISA modem (model SUP2124) in my
> > > Linux system running 2.4.20.  I've noticed that for the
> > > type of file I commonly access (large html text table) 115k
> > > port speed seems to be a significant bottleneck.  (These
> > > files compress to <10% of raw size.  Transfer speed is
> > > reported as ~10.5kByte/s, even if I reduce my modem speed
> > > from the typical 45kbps to 28.8.)  The modem documentation
> > > says it is capable of 230k port speed, so I tried specifying
> > > 230400 in my ppp config, which resulted in 9600 baud port
> > > speed; stty returns an error.
> > > 
> > > At this point in my education, my guess is that the modem is
> > > being treated like a 16550A without extended FIFOs or 230k
> > > port speed.  If my modem was detected as a modem by Linux,
> > > would setserial -a show something more?
> > > 
> > > # setserial -a /dev/ttyS3
> > > /dev/ttyS3, Line 3, UART: 16550A, Port: 0x02e8, IRQ: 10
> > >         Baud_base: 115200, close_delay: 50, divisor: 0
> > >         closing_wait: 3000
> > >         Flags: spd_normal skip_test
> > > 
> > > I am not a programmer, but have been looking at the comments
> > > in serial.c:
> > > 
> > > There seems to be an ISA PnP device device search, the success
> > > of which depends on finding the string "modem" in PnP id:
> > >   dev->name or dev->bus->name must contain "modem" substring;
> > > 
> > > However, the modem in this system was installed with PnP
> > > disabled and does not have modem in its string:
> > > # ANSI string -->SupraExpress 56i Voice<--
> > > 
> > > What are the consequences of failing this test?  If the
> > > string were changed, could the test succeed with the modem
> > > jumper not in PnP mode?
> > > 
> > > There seems to be a "quickie test to see how big the FIFO is".
> > > How can I cause the results of that test to be logged?
> > > 
> > > I now have CONFIG_SERIAL=y.
> > > If I compile serial as a module, are there parameters that
> > > can be added to modules.conf to enable available features
> > > beyond what a 16550A has, or perhaps someone has patches
> > > that would enable these features?
> > > 
> > > TIA,
> > > Tim.
> > > 
> > > -
> 			David Lawyer
> -
> To unsubscribe from this list: send the line "unsubscribe linux-serial" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 

port speeds >115k

[Tim]

[cc's not required]

> > > stty returns an error.
> > 
> > What error?  Quote exactly please.

stty -F /dev/ttyS3 230400
stty: /dev/ttyS3: unable to perform all requested operations

> > If the doco says it can run faster than 115200, it should also tell
> > what you have to do to make it do so.
> 
> Yes.  It will likely say (or imply) to use a driver they supply for
> Windows only.                                       ^^^^^^^^^^^

Ha! ~"All our 56k modems are capable of supporting 230k ... go find
a driver from someone else", actually.  Or try to ID the chipset &
find the datasheet.

> > > If my modem was detected as a modem by Linux,
> > > would setserial -a show something more?
> > 
> > It was detected as a serial port.  Linux does not detect modems at
> > startup.

I guess all that modem stuff in serial.c is just for PnP addr/irq init.

> > There seems to be a "quickie test to see how big the FIFO is".
> > How can I cause the results of that test to be logged?
> Is this test done by some Linux utility?

grep quickie drivers/char/serial.c

> > Internal modems usually just have a simulated uart for software
> > compatibility and baud rate is irrelevant (same for parity, stop bits,
> > FIFO, etc). Set to 5 baud and see if you get same performance.

Well I noticed stty <anyspeed> makes no dif in an xterm, or the 
getty on a vc, but with 
http://members.gate-way.net/delt/test/500k.html
my hardware ISA modem behaves like:

CARRIER	PORT	AVG.RATE
=======	====	========
46k	115k	10.5kB/s
26k	115k	10.5kB/s
14k	115k	10.5kB/s
46k	 57k	 5.3kB/s

For viewing large html text tables with a fast browser like Dillo,
230k could ~double the speed of the whole process.

> I set to 300 baud and couldn't connect.  Then at 1200 baud it connected but
> download speeds were around 60 bytes/sec.  I estimated the received rate
> from the /proc dir and it was close to 1200 baud.  So mine must have a
> real UART.
> 
> Some modems brag that they can do 230.4k baud.  This is a serial port
> speed.  So it seems like a real UART.  So what percent of internal
> modems will operate at a high baud rate when one specifies low baud rate?

I've seen non-rigorous test results on the net indicating that some
softmodems have no port-speed bottleneck.  It'd be great to have a GPL
HSP modem for Linux, now that GHz CPUs are the norm.  This groveling
for 230kbits/s through a device sitting directly on the system bus,
and being hobbled by hostile hw/sw producers is hard to take.  Wonder
what the absolute minimum hardware for a modem is...

Re: port speeds >115k

[whitnl73]

On Mon, 24 Feb 2003, Tim wrote:

...
>
> Well I noticed stty <anyspeed> makes no dif in an xterm, or the
> getty on a vc, but with

Right.  xterm and getty set the speed themselves, as any serial app can.
stty only matters if the app doesn't set the speed.

> http://members.gate-way.net/delt/test/500k.html
> my hardware ISA modem behaves like:
>
> CARRIER	PORT	AVG.RATE
> =======	====	========
> 46k	115k	10.5kB/s
> 26k	115k	10.5kB/s
> 14k	115k	10.5kB/s
> 46k	 57k	 5.3kB/s

This is ^ _bits_, not bytes, per second, and includes start and stop
bits.

...
Lawson
--
---oops---



________________________________________________________________
Sign Up for Juno Platinum Internet Access Today
Only $9.95 per month!
Visit www.juno.com

RE: port speeds >115k

[Ed Vance]

On Mon, Feb 24, 2003 at 9:58 AM, Tim wrote:
> 
> stty -F /dev/ttyS3 230400
> stty: /dev/ttyS3: unable to perform all requested operations
> 
> > > If the doco says it can run faster than 115200, it should 
> > > also tell what you have to do to make it do so.
> > 
> > Yes.  It will likely say (or imply) to use a driver 
> > they supply for Windows only.                                 
> > ^^^^^^^^^^^
> 
> Ha! ~"All our 56k modems are capable of supporting 230k ... go find
> a driver from someone else", actually.  Or try to ID the chipset &
> find the datasheet.
> 
> > > > If my modem was detected as a modem by Linux,
> > > > would setserial -a show something more?
> > > 
> > > It was detected as a serial port.  Linux does not detect modems at
> > > startup.
> 
> I guess all that modem stuff in serial.c is just for PnP 
> addr/irq init.
> 
> > > There seems to be a "quickie test to see how big the FIFO is".
> > > How can I cause the results of that test to be logged?
> > Is this test done by some Linux utility?
> 
> grep quickie drivers/char/serial.c
> 
> > > Internal modems usually just have a simulated uart for software
> > > compatibility and baud rate is irrelevant (same for 
> parity, stop bits,
> > > FIFO, etc). Set to 5 baud and see if you get same performance.
> 
> Well I noticed stty <anyspeed> makes no dif in an xterm, or the 
> getty on a vc, but with 
> http://members.gate-way.net/delt/test/500k.html
> my hardware ISA modem behaves like:
> 
> CARRIER	PORT	AVG.RATE
> =======	====	========
> 46k	115k	10.5kB/s
> 26k	115k	10.5kB/s
> 14k	115k	10.5kB/s
> 46k	 57k	 5.3kB/s
> 
> For viewing large html text tables with a fast browser like Dillo,
> 230k could ~double the speed of the whole process.
> 
> > I set to 300 baud and couldn't connect.  Then at 1200 baud 
> > it connected but download speeds were around 60 bytes/sec. 
> > I estimated the received rate from the /proc dir and it was 
> > close to 1200 baud.  So mine must have a real UART.
> > 
> > Some modems brag that they can do 230.4k baud.  This is a 
> > serial port speed.  So it seems like a real UART.  So what 
> > percent of internal modems will operate at a high baud rate 
> > when one specifies low baud rate?
> 
> [snip]

Hi Tim,

For a real 16550A UART, you cannot exceed the "baud_base" frequency, 
by calculating a new divisor, because the divisor would have to be 
less than one - and the 16550A can't do that. (16950 can) If the 
modem really supports a 230kb/sec UART rate, then there must either 
be a way to tell the card to either generate a higher speed base 
clock or change the UART's prescaler value. A real 16550A UART has 
a fixed prescaler that always divides the base clock by 16.

If you find a way to get a higher speed base clock, then use 
setserial to set the new baud_base value (must be 230kb or higher) 
and the warp flag that tells the serial driver to actually calculate 
a divisor for 230kb (spd_shi) when the user sets 38.4Kb (highest 
standard speed encoded in ye olde 4-bit termio baud rate field).

Ed

RE: port speeds >115k

[Ed Vance]

On Mon, Feb 24, 2003 at 11:20 AM, Steven Bradley wrote:
> Sorry to intrude, but ...
> 
> I suspect the interest in 230K is due to the 25 Mhz Courer USRobotics
> modems or similar species of modem.  Well, I would like to ask what is
> wrong with 115.2Kbps?
> 
> Let me be more specific with an example:
> 
> Assuming you have a 53Kbps connection, fastest available under 
> FCC current regulations limited power output, you would need a 
> throughput of 212Kbps if your data will compress 4:1 ratio 
> between computer and modem.  

I think 56kb is limited by the possibility of routing on a T-system
link that uses robbed-bit signaling. Anybody know for sure?

> In the earlier days before PPP and TCP/IP, data would 
> compress this much
> if it were certain combinations of ASCII or repetitive sequences of
> characters, such as binary empty sectors, for example.  
> 
> However, in practice, using TCP/IP over PPP, you rarely see this level
> of compression and 2:1 is more likely closer to what you will see.  If
> you are using 2:1, you would only need 106Kbps to achieve full
> efficiency because the mix of characters between modem and computer do
> not usually compress to 4:1 when using a PPP connection due to headers
> and control data which does not compress.
> 
> Even if you did have data which would benefit from the higher speed,
> the hardware flow control (CTS/RTS) would throttle it down for the
> RARE instances where this case may exist.  In most it would not.
> 
> Now that was assuming a full 53K connection, in practice, 44-50K are
> more common, assuming you have a 45K connection, it would only be
> 90Kbps throughput needed between computer and modem for a 2:1
> compressable data stream.
> 
> I thought this information may be of some interest, Steven

yep. Glad to hear from you. If 230kb has little throughput benefit 
over 115kb, I would use 115kb to minimize the number of data overruns.
I would be interested in hearing from anybody that has experience of
running at both digital side speeds with these modems under the
2.4 kernel. 

Cheers,
Ed

>  [ snip ]
> 
> -- 
> Steven Bradley / KG4UZR   121 Cambridge Drive, Longwood, FL  
> 32779-5707  USA
> steven@sgb.dyndns.org                 Home: (407) 862-7226 or 
> (407) 869-4598
>