csg@pyramid.UUCP (02/22/87)
I recently borrowed a pair of Racal-Vadic 9600VP modems to give them a work- out. The results were interesting, if not particularly exciting. BASIC STUFF: The 9600VP is a packetized 9600bps modem, conceptually similar to the Telebit Trailblazer although very different in implementation. Bell 103, Bell 212A, and the packetized mode (V.29) are supported. There is no V.22bis. Both the Hayes "AT" and traditional Vadic-style autodialing are supported. The high speed mode uses V.29 QAM encoding, half duplex, with a full speed of 9600bps and automatic fallback to 7200, 4800, and 2400. MNP is used for error correction, although it is Vadic's "Superset MNP," and includes compression. The 9600VP is incompatible with all other 9600bps modems, and will not commu- nicate with normal MNP modems even at slower speeds. High-speed full duplex is "emulated" using a technique Vadic calls Dynamic Du- plexing. 1200bps full duplex ("interactive mode") is used as long as the input buffers of both modems contain fewer than 150 characters. When this threshold is crossed, the modems switch to V.29 half-duplex ("file transfer mode") until the buffers are empty. Vadic recommends that the block size be a multiple of 3K. The time to switch carriers is described as long compared to normal Kermit packet size, but "inconsequential when factored over the required 3K block." The modem is packaged in Vadic's traditional wide, flat, chisel-shaped beige plastic cabinet, with a front panel composed of a row of nine LEDs and twelve illuminated membrame switches. The status LEDs include the usual TD, RD, DTR, DCD, and DSR, but no Off Hook. RI and CTS are also displayed, but RTS is not. A tri-color LED under the speed-select switch indicates the connect speed; the HS LED lights when V.29 is in use. An EC LED flashes when errors are corrected. The membrane switches are no where near as useful as in the RV212 modem; you can enable various test modes, set the speed, select sync/asynch, and switch between talk and data modes; but you can only dial numbers that were saved in memory, and cannot set the options. Of course, most modems have no front panel at all (e.g., Hayes, Courier, Telebit). There is no monitor speaker. Power is supplied by a 2x3x4inch brick with three feet of cable coming out of each end, so you don't have to hassle with the power brick covering two plugs on your outlet strip. The brick produces filtered, regulated DC, unlike most modems where the brick is just a step-down transformer and the power supply is inside the modem. A peek under the hood reveals a pair of Zilog Z80B CPUs running at 5.5Mhz, two 8K and one 2K CMOS SRAMs (the 2K gets standby power from a nicad cell), and a Rockwell piggyback board with an R5300, R5301, and 10462. (I thought that was Rockwell's 212A chip set -- anyone know for sure? The R5310, R5311, and 10464 are the V.22bis chipset.) Except for the CPUs and the line drivers, all chips are CMOS. Line switching is done by a rather loud mechanical relay. There is no cooling fan, which is touted by Vadic as a feature; the unit did seem to keep cool enough on my shelf. One possible trouble spot, though: the nicad cell is soldered to the board (typical :-( ) and it runs *hot*. Seems silly in a $1500 modem not to make that battery customer replacable. ADMINISTRIVIA: The test systems were both Pyramid 98x dual processors. pyramid, our mail/news server, was running its usual nightime mix of a half dozen uucp and news jobs, plus background test suites from the Cobol and Fortran crowd. pyrtech, the home machine for RTOC, was idle. vmstat(1) was running continuously on both machines to watch CPU load. A non-intrusive Halcyon datascope watched the data on the connection to pyramid. I don't have the necessary equipment to monitor phone line levels and such. I modified my "generic" Vadic dialer to select the 'f' protocol when the modem connected with an 'ONLINE 9600' message, and used my uucico that understands fixed interface speed. The modems were tested both with full autobauding and with speed fixed to 9600 bps. I did not test the AT command set, other than to verify that it worked. The 9600VP firmware is dated "Aug 28, 86", which Vadic told me was the latest version. (A newer version recently became available.) RUNNING THE TESTS: I had been told by the tech who evaluated the 9600VP for our Field Service department that the modems had not be able to reliably establish a connection. The Racal-Vadic rep who loaned him the modems blamed this on our Rolm CBX phone system. When I called Vadic, the rep claimed that this was not true; the only problem had been the tech's incorrectly setting the flow control. In fact, our two 9600VP's do not work through a Rolm CBX. The connection would usually "catch," but transfer was very slow, the front panel 'EC' light (error correction) would flash constantly, and the modems would finally disconnect after 20K bytes or so. For testing, then, I abandoned my usual outgoing phone lines through the CBX and went with a direct local Pa Bell line. That worked without a hitch. I queried Vadic about the CBX. After seeing for himself on the datascope that I was not botching the flow control, the Vadic rep claimed that no high-speed modem could be used over a CBX because of the digitizing process it uses. I immediately pointed out that our Trailblazers regularly shovel data through at 1050cps. After a pregnant pause, he decided to take back both of the loaner modems, and promised he would take them to ROLM for analysis, and return them when they worked over the CBX. That was about a month ago. Peter Holzmann at Octopus Enterprises was kind enough to dredge up information about the CBX and QAM encoding. It appears that a properly set up digital PBX should have no trouble passing the Vadic's 9600bps QAM. The problem would thus have to be in line levels, line balancing (could cause echos), or something similarly mundane. RESULTS: Interactive was no contest: the 9600VP was *very* pleasant to use, with none of the annoying pauses that plague the Telebit. Keystrokes were always echoed promptly, and full screen refreshes came in a rush. For interactive terminal work the dynamic duplexing seems to do its job. I did notice regular pauses during very large runs of output (like ls -l /usr/bin), but they were not bothersome. The Vadic rep said those were for "retraining." According to the manual, the modem drops back to full-duplex 1200 every 3K bytes to exchange ECC packets. One serious problem: flow control is enabled only when the modem is running V.29. So if fixed interface speed is used (9600bps) for a 212A connection, it is very easy to overflow the buffers. For this and other reasons, the 9600VP should be avoided for normal Bell 212A or 103 calls. UUCP 'g'-protocol was worse than awful. 70 to 100cps, with lots of retries due to mangled packets. The modem never saw a big enough packet for it to get out of interactive mode, which I expected, but the bad transfers worry me. I saw nothing on the datascope that made it clear what was happening. It may be due to the lack of flow control in interactive mode, or some weird boundary condition switching between interactive and fast mode. Using the 'f'-protocol, I got a consistent 718 characters per second (after subtracting out the 'f' overhead). And I mean *consistent*. You could have used it for a stopwatch; very different from the Telebit, which constantly varies the data rate. The datascope confirmed that transfer was occurring in bursts at 9600bps, with the regular pauses for "retraining." If compression was involved, it worked well enough that I could not see it. Flow control technique didn't seem to matter; the Pyramid and the 9600VP were equally happy with XON/XOFF or CTS/RTS. The loaded Pyramid had a lot of slop on its XOFF response, too; up to 24 characters would trickle out. The 9600VP always stopped within 3 characters of XOFF. I don't have SLIP available, so I could not test that. Mark Weiser at Univer- sity of Maryland has reported very good performace, with a real data rate (ftp and rcp) of about 690 cps. Telnet was terrible, however, since the big TCP/IP packet headers confuse the dynamic duplexing algorithm. Impact of the 9600VP on the system was surprising, less than 1% CPU load. This is actually less than that of a normal 2400bps modem using 'g'-protocol (3%), and much less than that of the Telebit running 'f'-protocol (8%). I have to believe that this was a fluke of the 9600VP and Pyramid SIO combination; I sure wouldn't expect to see that kind of results on a DH-11. MISCELLANEA: Interface autobauding worked well, which is an advantage over the Telebit. Good thing, too, since you really can't use fixed interface speed because of the lack of flow control. Call progress monitoring worked well, although not as well as on the Telebit or on the better 2400bps modems (e.g., AT&T 2224). As usual, all the option numbers are different from all previous RV modems. The help menu lists all the commands and options, but not what their arguments or values mean. The Trailblazer suffers from the same half-hearted menus. The DCD lines can be set to RS-232C, always on, or Vadic's "Computer Systems Interface." The DCD line is asserted until the modem attempts to dial a call. DCD then falls until a carrier is received; it then is asserted until the true carrier falls. I could get the uucico Vadic dialer to use this instead of forcing DCD always on, but I didn't think it was worth the effort. (I prefer Microcom's design, where DCD pulses low at the off transition of the real carrier.) 212A calling other modems was disappointing, definitely below par. The 9600VP seemed to be very vulnerable to line noise, and often failed to connect. SUMMARY: Like the Telebit Trailblazer, the Racal Vadic 9600VP is a bag of plusses and minuses. It does what it's supposed to do pretty well; so does the Telebit. Both can pay for themselves in telephone bills, and are faster then PDNs. Both are a pain to use with existing software. Line quality was perfect, so I'm not surprised that I never saw fallback. I have no idea how the 9600VP would perform over noisy lines, although its be- havior on the CBX worries me. I never saw it pass bad data, however, something the Trailblazer is known to do when line quality is exceptionally poor. Plusses are: excellent interactive response; comparatively low cost ($1500 list, $1050 to educational institutions); simple design (should be reliable); some handy front panel switches. Minuses are: No V.22bis; poor 212A; no monitor speaker; overly sensitive to line characteristics; a trifle slower than a Trailblazer on perfect lines. If it were my money, I'd be very reluctant to buy either the 9600VP or the Telebit Trailblazer right now. The 9600VP is the clear winner for interactive high speed, but the Trailblazer has a very good V.22bis (which we use a lot here), and an slightly superior file-transfer capability. Both seem to pale next to the crop of 9600/1200 reverse channel modems that are now available. <csg>
butler@stsci.UUCP (02/26/87)
Last November I had to evaluate several high speed modems for use in some networking applications. I have always meant to do a full writeup and post to the net, but have not had time. My findings at the time were that the RACAL-VADIC was more pleasant as an interactive use modem than the Telebit, and was a bit cheaper. A modem that was of interest but which arrived late for me to do much testing and evaluation was the ADDCOM. It looked to have some of the features the VADIC lacked (speaker, line quality indicator). As far as their use, we are using them to implement some phone links while we wait for DDS lines from the Death Star company. They don't show us the throughput I wish they did (Still no comparison to a 9.6k DDS Line) but when the packets get big, the modems push 'em through pretty good. Throughput seems to be more on the order of 7200 than 9600 (because of the resync'ing the modem does I guess). Telnet or "set host" are not the nicest things, but file transfer isn't as bad as over 1200 baud modems by a long shot. I have NEVER seen a bad character come out of the modem. I HAVE seen it hang up due to line noise. We had some real problems with our phone switch and these modems. We ended up getting direct lines from the local carrier for these modems. Seems with all the error correction and compression going on it takes a moment for the modem to realize the phone has been hung up. Our phone switch generates a "Busy" signal when the remote party hangs up. The modem kept trying to recognize a carrier in the "busy" signal. I'm not sure whether to blame the modem or the phone switch. The Telebit had a similar problem with the phone switch. There has been some difficulty getting trans-continental phone calls to connect at something above 1200 baud. It seems to depend on solar activity, lunar position and Nancy Reagan's mood. Perhaps I'll post more later if there is further interest on this subject. Alternatively, interested parties should feel free to contact me at one of the addresses below. Lee A. Butler Space Telescope Science Institute 3700 San Martin Dr. Baltimore MD 21218 Arpanet: butler@stsci.arpa | butler@brl.arpa Usenet: seismo!stsci.arpa!butler | {noao,astrovax,cfa,charm,nrao1}!stsci!butler SPAN: scivax::butler (6405::butler) Phone: (301) 338-4531 -- Lee A. Butler Space Telescope Science Institute 3700 San Martin Dr. Baltimore MD 21218 Arpanet: butler@stsci.arpa | butler@brl.arpa Usenet: seismo!stsci.arpa!butler | {noao,astrovax,cfa,charm,nrao1}!stsci!butler Phone: (301) 338-4531