BEC.SHAPIN@usc-ecl@sri-unix.UUCP (10/05/83)
From: Ted Shapin <BEC.SHAPIN@usc-ecl> In regard to the Aug 14 message of Richard Foulk, Honolulu regarding a faster protocol, the following contribution is from Harold Price. - - - - - - - Thoughts on a faster transfer protocol than the Christensen protocol. First, increasing the frame size will increase thruput only if the bit error rate of the transmission medium is low enough. Obviously, not only does a 1k frame take longer to resend, but it is a bigger target for a line hit. Error rate aside, larger frames is the only way to increase thruput as long as two other things remain unchanged, the line is treated as half duplex, and only one frame is sent at a time. If more than one frame is sent at once, the number of turnarounds can be reduced. A fancy error recovery scheme can take care of retransmission of damaged frames. The number of frames sent before line turnaround depends on the memory buffer size of the destination machine, assuming that the destination can't write to disk and read the data at the same time, and on the error rate of the transmission line. A line with no errors and a destination with infinite memory would require no turnarounds. The problem of how many frames to send is solved if the line can be treated as full duplex. In this mode, frame n is acknowledged while frame n+1 is being transmitted. This is very efficient, the only overhead is the frame number and checksum. This method would also promote simultaneous two way data transfer, the acknowledgment of an incoming frame can be piggy-backed on the next outgoing frame. Suprisingly enough, such a scheme has already been defined. In general, such a thing is called a sliding window protocol. A good example of one is LAPB, a protocol which forms the basis for level 2 of the X.25 network standard. LAPB can be used as the transmission protocol for an efficient two-way file transfer scheme. It is somewhat more difficult to implement than the MODEM7 protocol which has grown out of Christensen's work, but not too much so. Several implementations of LAPB for micro computers have been done as part of the growing "packet radio" movement in Amateur Radio. The protocol is called AX.25, since it is modeled on X.25. AX.25 is the protocol which is currently recommended for use in amateur terrestrial networks as well as space based transmisson modes using the recently launched OSCAR-10 high earth orbit satellite. AX.25 will also be used for PACSAT, an orbiting mailbox system with 4MB of on-board memory, due to be launched in 1986. AX.25 implementations exist for 6809, Z80, and 8085 systems, several of these are implemented in C or Pascal. They are not currently configured as "file transfer" programs, but as more generalized "data transfer" processors. A high level file transfer module can easily be added. To move on to the subject of tampering with a defacto standard, there are now enough "commercial" file transfer packages around using the Christensen protocol that it is unlikely to die. Because of the off the shelf nature of these packages, most users can't or won't tweak their program or add features to it. If they use a new protocol it will be because they bought it or someone gave it to them. My point is as long as you've got to do a complete replacement anyway, don't limit yourself to thinking in terms of a slight modification to the existing MODEM programs. Go the whole way and get the most out of your connect time, use a sliding window protocol with simultaneous two-way capability. Drop a working model off at each RCPM system you encounter and encourage the SYSOP to make it available on his system. There are, or course, commercial packages available that use sliding window full duplex protocols, and other efficient protocols, so if you have enough $$ you can buy these for the systems you interact with most. I would recommend looking into the AX.25 work that has already been done. Most of it is in the public domain. For more info you can contact me at the address below, or any of the amateur radio groups working on AX.25, ARRL, AMSAT, TAPR. The addresses of these organizations can be found in any of the more recent amateur magazines, QST, 73, and Ham Radio. Harold Price, NK6K PACSAT Project Manager 1211 Ford Ave Redondo Beach, CA 90278 -------
rjh@ihuxh.UUCP (Randolph J. Herber) (10/19/83)
You may want also to look into HASP (IBM OS/360) Multi-leaving protocol. It has similiar capacities to piggy-back responses and handle two-way data flow. And, it could ease communication with IBM-compatible mainframes. Randolph J. Herber, ..!ihnp4!ihuxh!rjh.