lamy@sobeco.com (Jean-Francois Lamy) (09/01/90)
I'm staring at an ad in {Communications Week} about the Motorola UDS
FastTalk V.32/42b modem that is claimed to provide throughputs of up
to 38.4. We have an application currently using stat muxes over a
leased digital 19.2 line that we'd like to move to TCP/IP, and
compressed SL/IP (or PPP) seem like good candidates for the
application.
We suspect however that the additional demand created by the added
functionality and extra overhead will spell the death of the 19.2
connection. Leasing 56kbps lines in Canada is outrageously expensive
(8000$ a month), and we don't have enough voice traffic either to
justify going to fractional T1. We've seen boxes at 7000$ each that
do quite a good job muxing data from four channels at 38.4,
compressing down to a 19.2 sync link -- we're using only one channel
at 38.4. Sending pre-compressed data, however, reduces actual
throughput to close to the line capacity, as one might expect
(fortunately the unit does not actually degrade in those cases). But
14000$ for 2 units, plus the cost of modems and the line is a bit
much.
We need sustainable 19.2 at the very minimum, and more the better.
Has anyone tried the Motorola UDS FastTalk V.32/V.42b and seen whether
the unit will --
- maintain 19.2 in the presence of pre-compressed (meaning compress(1)
Lempel- Ziv) data such as a bozo transferring a tape over a serial
line, or will it sink down to its "native" 9600bps?
- achieve anything close to 38.4 doing SLIP or PPP. (Where what is
flying by are either very short packets (2-10 chars) or short packets
(we'd be keeping our MTU size down to keep echo latency down).
I will summarize replies mailed to me. Vendor plugs welcome too, to
my e-mailbox, that is.
Jean-Francois Lamy lamy@sobeco.com, uunet!sobeco!lamy
Groupe Sobeco, 505 ouest, bd Rene-Levesque, Montreal Canada H2Z 1Y7BRUCE@ccavax.camb.com (Barton F. Bruce) (09/04/90)
In article <11625@accuvax.nwu.edu>, lamy@sobeco.com (Jean-Francois Lamy) writes: > FastTalk V.32/42b modem that is claimed to provide throughputs of up > to 38.4. We have an application currently using stat muxes over a You don't want a V.32/V.42bis modem, you want a V.32bis/V.42bis! The dust may not quite have settled in the standards arena, but IF you buy a product from a company that is doing the modem part of it with DSP technology (as opposed to having used the Rockwell chips), it is quite reasonable to assume they can cover all bases NOW, and if push comes to shove and someone fiddles yet again with the standards, they can simply give you a new prom or (for those that blast their own) let you suck it off their BBS. V.32bis does 14.4 FULL DUPLEX as its native speed. The V.42bis gets you the error correction and compression you want. Having V.42bis on a 9.6kb V.32 modem is 'nice', but why not start with the latest pre-compression speed. With normal success for compression these days, 19.2 just isn't enough, you need at least 38.4, especially if you are starting at 14.4. You may well have other problems, brought on by this speed. If you have a built-in PC card modem, pray they either use a UART like National's NS16550AN, or at least have it socketed so you can stuff one in. If you use an external modem, most, even the 'el-cheapo', AT-IO cards have socketed UARTS, and many simply leave the 2nd one unpopulated, anyway, ready for you to stuff in an NS16550AN. Someone who knows for sure better correct me, but I think any standards level bickering about v.42bis is currently over what other lower speeds are supported, not about the 14.4 itself. A company like Digicom Systems Inc (DSI) right NOW will give you V.32/V.42bis or V.32bis/V.42, but not both. The current hardware should be readily upgradable is a few WEEKS to the new proms to do V.32bis/V.42bis! Their "PLUS OPTION" upgrade package will take a (not so) vanilla v.32 modem listing at around $795. and make it do BOTH V.32bis AND V.42bis as well as to do Group-3 FAX! (they include a floppy for for their MSDOS 'FlashFax' s/w - the OS2 and Unix flavors are being developed) is about $200. more list. Of course noone ever pays list ... That was the boxed modem price, the PC card will be less. If you want to take T1 in, there is a company that makes a rack mount little gem that can take in up to 20 T1s! and on DSP cards that can start you at economical v.22(bis) type speeds. When you need faster speeds for any of the 480 modems that implements (20 x 24 = 480), you just pay for new license to run the V.32 (or better) s/w! You can get all 480 ports out as eia cables if you really want, but you may prefer to let them keep the signals inside and run them through the built in PAD and bring all out via a HIGH SPEED x.25 cable. These can be used for central computer sites, but, since they also work on feature group B and D trunks, anyone with enough traffic can deploy these around the country where 'normal' IXC carriers would connect to LECs, and get user's dial in traffic on 950-xxxx (feature group B) or other creative numbers (but needing 10xxx or default status) on feature group D trunks. WHY? Try saving probably OVER 1/2 the cost of using 1-800 terminating into T1s. If you are a credit card verification, or Compuserve, or a Telenet, or Amex or <unmaned Mega-Bank> type application, this is where your world probably is heading. Given time, this hardware might just implement LAT/TCP-IP terminal server functionality as am alternate to the built in PAD. Lest I forget, for the few of you that need something this big, they are: Primary Access. That slight digression was simply to point out that DSP chips right NOW are in modems that are UPGRADEABLE by S/W (firmware if you prefer to nitpick), and that you can BUY today. If you can't get v.32bis with v.42bis, or a FIRM no/low cost upgrade commitment for the very immediate future, try another brand! - no connection to any vendor above (yet)...