NetNews Offline 2/NetNews Offline Volume 2.iso/news/comp/dcom/modems-part2/11549

Internet Message Format | 1996-08-05 | 8.7 KB

Path: FreeNet.Carleton.CA!an171
From: [e-mail] (Anthony Hill)
Newsgroups: comp.dcom.modems
Subject: Re: (E)nhanced (S)erial (P)ort...
Date: 19 Apr 1996 05:00:17 GMT
Organization: The National Capital FreeNet
Sender: [e-mail] (Anthony Hill)
Message-ID: <4l76l1$[e-mail]>
References: <4l6i6p$[e-mail]>
Reply-To: [e-mail] (Anthony Hill)
NNTP-Posting-Host: freenet5.carleton.ca


Ed Starry ([e-mail]) writes:
>     I've called this a cached serial port with caching controller to
> differentiate from the 16-byte 16550 UART 'Buffer' to avoid confusion in
> using the word buffer. John N. calls them 'smart buffers', however smart
> buffers need controllers to make them smart, a buffering controller. Buffers
> cache and caches buffer! Same thing, different language. "Lets call them
> what they are, (E)nhanced (S)erial (P)ort!"

    A buffer is memory that temporarily stores data until the
processor (or whatever) can handle it. A cache is memory that stores data
that is frequantly accessed, and since memory can be accessed much faster
then hard drives or CD-ROMs or anything like that, a cache appears to
speed accessing data from these devices.  16550 UARTs use a 16 byte
buffer, 16650 UARTs use 32 byte buffers, 16750 UARTs use 64 byte buffers,
Hayes ESP IIs use 1024 byte buffers.

>     There are single-byte serial ports (8250's), multi-byte serial ports
> (16550's) and the next step up the ladder, ESP's. When I made the comment a
> few days ago regarding 'If people would start demanding (and using) 'cached
> serial ports' they would discover even 14.4's are much faster than they
> realize.' these ESP cards were what I was referring to.
>
>     An external modem plugged into an ESP card is similar to putting good gas in
> your car instead of kerosene, they run as designed. These cards with proper
> configuring software allow your external modem to run at the factory defined
> DTE rate. e.g., 14.4's@57,600 or 115,200 - 28.8's@115,200 or 230,400 and
> 33.6's@134,400. Quality internal modems have these ESP's built in, that's
> why I've been able to run my 14.4@115,200 without problems for the last 3
> years. Admittedly, TurboCom/2 was used to replace the default <comm.drv> in
> Windows V3.1 as this default driver is truly substandard for today's high
> speed modems. If I'd been using the same make and model of modem only an
> external version I would be running at 38,400 and fighting overruns like
> everyone else. Even a 16550 by itself will not 'fully support' 115,200 bps,

    Shhh!  Dont' tell that to my 16550!  It runs at 115.2kbps just
fine.  No com overruns or anythign like that..

> despite what John says. "An ESP doesn't guarantee full DTE rates because
> there is always 'The Other Modem' but you will know any bottleneck there is
> is not your modem!"
>
>     I know this post will probably raise more flack directed toward me but press
> on I will. If everyone was using ESP's then ISP's and BBS's might stop using
> this ludicrous '1 byte per IRQ' upload rate they now use. To keep from

    What the heck kind of ISP are you calling that connects their
modems to real serial ports?!?  Just about every ISP I've seen uses a
terminal server, or at the very least some sort of multiport serial card
such as a Digiboard.  This has nothing to do about UARTs, it's simply
because you can only have so many serial ports in a computer without
special hardware (in the case of a PC, the maximum number of serial ports
you can have is 4).  Also, 16550s can set their Rx trigger level to either
1 byte, 4 bytes, 8 bytes of 14 bytes (I think I've got those numbers
right).

> crashing every client they have they have no choice now. The first day they
> tried to run wide open there would be crashing like never heard before.
> 28.8's running at 57,600 would be 'blown out of the water' in 10 seconds, or
> less, without an ESP. I have a personal interest in this, I've experienced

    I commonly connect to BBSes that have their DTE rate set at
115.2kbps with no problems.. They're all using plain old 16550 UARTs.  As
for ISPs using Hayes ESP cards, uhh, yeah.. whatever.  I think I'll just
call up my local ISP with 600+ modems and tell them to dump all their
terminal servers and buy 150 PCs equipment with 4 Hayes ESP cards each..
I'm sure they'll go for it :>

> fast and I want my ISP to crank things up to screaming fast. Near ISDN rates
> from a 14.4 would save me major money in the long run.

    Same here, and if you ever figure out a way to get anywhere close
to that, you'll be a very rich person.. Unfortuantly you won't (hate to be
blunt, but it's the truth).

>     Now for two little items plagiarized from magazines.
>
> 1.    From PC Magazine, May 14, 1996 - page #78. 'Need more speed? You can
> make faster connections with Lava. This interface card is designed to
> increase the throughput of ANY V.34 analog modem. It also will allow ISDN
> users to achieve real 128 Kbps PPP throughput from an external terminal
> adapter.' This 'was NOT' an advertisement.

    So what is it?  A synchronous serial card?

> 2.    From a BitSURFER Pro ISDN advertisement. 'External model requires
> serial port interface capable of 128 Kbps to achieve maximum performance.'

    Yup, and guess what?  You're Hayes ESP card does NOT support that
128kbps needed for maximum throughput :>.  The highest speed that the
Bitsurfer Pro will run asynchronously is 115.2kbps.  To run at 128kbps, it
must run in synchronous mode, and ESP cards don't support synch rates.  So
basically in this case (as with most modems) you're ESP can NOT run the
device at a higher speed then a regular 16550 UART.

>     A 33.6 should run at 134.4 Kbps. Is your 33.6 running at 13.44 KByte/sec?

    Considering I transfer mainly compressed data, my 33.6 would
see next to no perfomance increase if it ran at 230.4kbps (134.4kbps is a
not a particuarly valid asynchronous DTE rate) vs. 57.6kbps.

> Does anyone see a correlation here between this and my earlier statements
> regarding caching/buffering of the serial port? Gee, they even let a plain
> old V.34 run faster, what an enlightenment! I've read where the 'experts(?)'
> say a 16550 is all that's needed and ESP serial ports were unnecessary!!
> Could these experts be wrong?

    They could be, but they aren't :>.

>     I will stand by my statement regarding 'marketing ploy' until John, and
> others, realize there's a feature of modems called 'Compression', encourage
> the usage of and teach people how it's done instead of dismissing it as

    Ed, please drop into DOS, compress a nice big file (1 meg or more
preferably) and compress it with PKZIP.  Then compress it again using any
compression program (even PKZIP again).  Now compare the sizes of these
files.  You will notice that you can NOT decreas the size of the file by
compressing it twice.  Now, how do you expect your modems compression
(which isn't as good as PKZIPs) to compress something twice if no other
compression program will?  Since most people transfer mainly compresed
data, compression doesn't play all that much of a role.  As for text files
and other forms of uncompressed data, try compressign these files using
PKZIP and see what sort of compression ratios you get.  About the best
you'l manage with most real world files is 4:1 with PKZIP.  As I mentioned
above, your modems compression is not as good as PKZIPs (hey, it's got to
run in real time, with very little memory and on a processor that isn't
too powerful).

> 'that's all you should expect'. I expect DTE rates equal to, or higher than,
> manufacturers' specifications and nothing less. "I paid for it and I want
> it!! Passivity is no longer acceptable!"
>
>     An interesting side effect with connections where *BOTH* modems are using
> properly configured ESP's is what happens to zip/gif files. You will have to
> find this secret out on your own. <g>

    It's not a big secret, there is NO difference in performance.
Please, try it for yourself.  Transfer .ZIP files between two modems using
Hayes ESP cards, then try again between two modems connected through 16550
UARTs (or just disable the extended functions on your ESP cards).  Unless
you're getting com overruns with the 16550s, there will be no difference
in performance, and unless you're running Windows you won't have any com
overruns.

> PS:    An 'Enhanced V.42bis' modem is a modem that compresses at a 8 to 1
> ratio and they require an (E)nhanced (S)erial (P)ort to run at full speed.
> Yes 8 to 1 compressing modems, contrary to what others tell you, do exist
> and they will work as designed if set up properly.

    Sure, and if you transfer a file filled with all 1 character over
and over again you might actually see those sorts of compression numbers.
BTW, my modem can compress data in ratios of up to 96 to 1 with a 16550
UART.

Anthony

--
Anthony Hill | [e-mail]