Newsgroups: comp.dcom.modems Path: utzoo!utgpu!cunews!hobbit.gandalf.ca!dcarr From: dcarr@hobbit.gandalf.ca (Dave Carr) Subject: Re: USR Courier V.32bis questions etc.... Message-ID: <1991Apr21.185752.2200@hobbit.gandalf.ca> Organization: Gandalf Data Ltd. References: <6465@husc6.harvard.edu> <1991Apr21.032236.21724@netcom.COM> Distribution: na Date: Sun, 21 Apr 1991 18:57:52 GMT Lines: 36 In cs352a41@cs.iastate.edu (Adam Goldberg) writes: >>I think the reason most modem manufacturers are creating modems with a max >>RS232 speed of 38400 are doing it for three reasons: (a) It doesn't require >>exotic, costly processors, (b) Real life data transfers won't be more than >>2.7:1 compressible 98% of the time, and (c) Few computers can handle data >>rates above 38400 bps. In short, they're designing a modem that will meet >>your data transfer needs without exceeding your budget needs. (D) RS-232 has its limits. You wouldn't want to advertise you can do 56 or 64K over RS-232, otherwise the end-user starts screaming at your technical support people that he is getting errors on his "short" 10 foot cable. >Since it is (theoretically) possible for an effective data transfer rate of >57,600, but it would take 'exotic, costly processors', it struck me: Hmmm, >is it possible (or has anyone attempted) to: Exotic, no. Current, Yes. One dedicated 80960CA ($75US), can COMPRESS a 256K bps link using V.42 bis compression. Compressing a 14.4 K bps link is almost archaic. > o Construct such a very, very fast modem. > o Design it as an internal card with DMA--ie, it would take a very strange > RS232 port to support such high speeds, and the processor would be > completely taken over by it, so wouldn't it make sense to have the modem > work like a disk drive and deposit data directly to memory? Clearly if > a HST modem has a 68k inside of it... No need to get this fancy at this speed. Just keep the modem cable short. RS-232 should :-) work at 56K for 2 feet or so. No need for DMA. Just deeper FIFO's on the USART. Put a Z16C30 USC with 32 byte FIFO's, and the service routine would only need to run every 20 milliseconds. CPU overhead lower than most serial ports use at 9600.