wtm@neoucom.UUCP (Bill Mayhew) (07/30/88)
Hi, We just received two Trailblazer Plus modems today to try out on loan. We've been using the older version 3 Trailblazers between two of our internal sites, spl1 and pitt for a while. The Trailblazer Pluses have dropped the rock-like battleship gray strutural foam case for a thin hard styrene plastic case. There was something about the rock-like feel of the older case that seemed to imply that the Trailblazer was a modem that meant business. Fortunately, the new model still seems to deliver on the promise. Gone too is the small fan. The fan really seemed superfluous in the old version, and the Plus seems to be quite cool without it. The case is only slightly warm to the touch. The innards of the Plus seem fairly similar to the "classic" version of the Trailblazer. The TI DSP chip, Motoroal 68K and AMD Worldchip are all still there. Much of the 74LS series logic seems to have been aglomerated into a big ASIC. Three small daugherboard assemblies, two bearing RAM and one analog, have been droped in favor of more conventional components in the Plus. At first, it would appear to be a step in the reverse technological direction, but I feel it makes the modem simpler to manufacture, and probably more reliable with everything on one PC board. The voltage regulator circuitry has been substantially simplified and generates much less heat under operation. Last of all, the PC board layout has been revamped substantially, apparently to cut RF emissions. Emission without the phone-cord torroidal coil and without metal-lined case seems low. My qualitative test is that I can still operate my cordless phone nearby. The Trailblazer Plus does sport two pushbuttons on the front panel now. One button can be used to select from one of two possible sets of nonvolitile register configurations. The other button is marked T/D and can be used to toggle between voice/data or to manually hang up the modem. The small reset button hole on the rear panel has disappeared, seemingly replaced by an IBM MS-DOS setup floppy disk. Since neither of the two machines in the test below runs MS-DOS, I can't speculate on how helpful the disk is in configuring the modem. We just used the Unix operating system's cu program for setting up. We set up some test files on my system, impulse, and the main node, neoucom. We used the termcap file glued together several times to come up with a 154780 byte text file. Impulse is an AT&T 3b1 running HoneyDANBER uucp and a System V-like operating system that AT&T calls "release 3.51". We attached the Trailblazer Plus to the 3b1's internal tty000 serial port, and configured the Systems and Devices files for 19200 buad. AT&T's manuals state that the serial port does not operate beyond 9600 buad, however HDB seemed to accept 19.2K without complaint. I'm sure that it was indeed 19.2K, as the Trailblazer was set to have register S51=5 (interface locked to 19.2K). The machine at the other end is a Vax 11/750 running Berkeley Unix operating system version 4.2. We conducted the test will eleven users were logged in with loads of 0.5 to 2.6. We tested the Tailblazer on both a DZ-11 8 port board and an Emulex DH-11. Both the the DH and DZ have some built-in FIFO registers to improve throughput. The DH-11 supports 19.2K, but our DH lacks the option to suuport full modem control, thus it is not possible for us to test the Trailblazer with uugetty on the Vax. The Vax was also running a very eraly incarnation of HoneyDANBER uucp. Curiously, we found that the Vax could transmit characters throuh the Trailblazer faster than it could receive, while on the 3b1 using the same register contents, it seemed to make no difference. I suspect that the Vax's kernel could have used some tuning to add a few more disk buffers. The phone line between impulse and neoucom is normally incapable of supporting a normal 1200 baud dial-up becuase of continuout phase hits and a high frequency impairment. The adaptive equalizer in the Trailblazer was able to clean-up most of the problems so that a more-or-less error-free dial-up was possible. In the proprietary "PEP" high speed mode, errors disappeared completely. The modems' data registers showed a peak instantaneous baud rate of 16400, which may or may not be representative of the real world. In interactive sessions, there is a turnaround dealy of apparently around 40 mS, as the modem changes transmission directions. Since the Trailblazer mediates uucp g protocol, xmodem or kermit packets internally, the fact that the PEP is actually half-duplex is mitigated, and transmission throughputs approaching the instantaneous maximum can be obtained. We found that transmission rates varied between 313 and 1251 (!!!) characters/sec on the same 154780 byte file. Such a wide range of rates seems to be due to the load factor on the Vax. We ran uucico with a nice value of -14 to try to enhance throughput. Nice'ing seemed to be especially important while the Vax was receiving characters. Impulse didn't seem to care, sending and receiving characters as fast as the modem could deliver. It is clear that on our Vax, the Trailblzer really only reaches its maximum potentail when the machine is lightly loaded. The 154780 byte test file was a flat ASCII file, and thus benefited from the Trailblazer's compression. Usenet batches would probably be best sent with modem compression disabled, as the second application of compression is superfluous and likely to slow down transmission slightly. We had been running news batches between neoucom at 2:00 am, and getting about 620 char/sec with the Trailblazer running at 9600 baud. We have been getting the above quite consistently for several weeks. Today is the first day that we've been running at 19.2K, so it will be interesting to see how the news feeds do. I hate to sound like an adverstisement, but that figure of 1251 characters/sec is pretty amazing. That was on a pretty big file, and usr/spool/uucp/.Admin/xferstats showed the same figure on both sides, so I think the effect of the modem's internal buffer was not unduly slanting the figure. Bill Bill Mayhew Electrical Engineering Division of Basic Medical Sciences NE Ohio Uviversities' College of Medicine ...!spl!neoucom!wtm ...!pitt!neoucom!wtm ...!allegra!neoucom!wtm Our domain is in a state of flux, thus the current net map may not be accurate for our site's connections. I apologize in advance for any bounced mail.