Using SIXNET Universal Protocol over Ethernet TCP/IP     Last Rev: 10/13/98

SIXNET Ethernet I/O stations provide a minimal TCP/IP protocol suite
implementation to allow a standard TCP/IP protocol stack to communicate
with it. All SIXNET Ethernet I/O and controllers share this common basic 
implementation and you may expect that drivers developed to support SIXNET 
Ethernet networking will work with all applicable products without exception.

SIXNET Ethernet I/O and controllers support the ARP, IP, ICMP and UDP protocols.
UDP is used to communicate with the station. ARP, IP, and ICMP are used to
support features of the TCP/IP facilities of the host operating system, so
that it can determine the Ethernet address of the SIXNET station given the
IP address, send UDP messages, the ping utility can see the station and so
on. They are normally invisible to the program using UDP to communicate to
SIXNET. The required minimum IP packet size of 576 bytes is supported.

UDP port number 1594 is the Registered Port for SIXNET Universal Protocol
communications. This is the default port for SIXNET Ethernet stations. In
the future it may be made configurable, but SIXNET software does not
presently allow this to be set by the user. A SIXNET Ethernet station accepts
messages from any UDP port and IP address that are sent to UDP port 1594
and its IP address. The reply is sent from UDP port 1594 and its IP address
to the originator's UDP port and IP address.

Standard SIXNET Universal Protocol messages are used to communicate with
SIXTNET stations. Since the IP addresses uniquely identify the source and
destination, the SIXNET Universal Protocol source and destination numbers
could each be set to `FF'hex, (a special one byte address meaning
`any station'), even with a multi-station and multi-master configuration.
If a specific destination number is used (other than `FF'hex) in the
message it must match the station's number or no reply is generated.

SIXNET Universal Protocol Binary Format and Fixed CRC messages are
supported. Hex format is not supported since 8-bit data is always supported
on Ethernet, and Hex conversion wastes a lot of processing time to encode
and decode at both ends of the communication. This wasted time becomes a
significant burden on the response time, given the high speed of Ethernet
communications.

The Fixed CRC format saves even more processing time at each end. It is
safe to use when the path is known to include error checking at least as
good as the 16-bit CRC used with Binary Format. For example, if a computer
is hooked to the same Ethernet network as the SIXNET station, the 32
bit CRC computed on the Ethernet packet provides a sufficient error check
and wastes no processing time since it is computed and checked by the
Ethernet hardware.

One or more complete SIXNET Universal Protocol messages are sent as a
single UDP message. The data portion of the UDP message (the SIXNET
Universal Protocol messages) must be no more than 508 bytes. This makes
sure that the IP packet is not more than 576 bytes (60 for maximum IP
header + 8 for UDP header + 508 data). If more than one message is sent
in a single UDP packet the sum of the length bytes + 6 bytes/message must
not be greater than 508. To be sure that all replies will fit in a single
UDP reply the replies must satisfy the same rule. Watch out for the GETS
command, since the length byte of the reply is unknown in advance, and
could be as large as 255! If the replies to the messages in a single UDP
message exceed 508 bytes multiple UDP messages are generated. If multiple
session numbers are referenced in a single UDP message the replies will
not necessarily be grouped in a single UDP message, even if they would
fit.

UDP Reliability

In some published documentation, UDP is sometimes described as
'unreliable', while TCP is 'reliable'. This is somewhat misleading, since
in both cases the data travels as an IP packet in a similar manner with
identical reliability. The IP packet receives the same delivery processing
in either case, so it has the same chances of arriving at the destination
independent of the higher level protocol being used. In the case of two
nodes on an Ethernet network this is basically 100%. (Halfway around the
world through the Internet, it may be somewhat less).

The difference between TCP and UDP is who handles retries if a message is
lost or duplicated. The TCP transport layers handle this for the client
program, while UDP does not. The SIXNET Universal Protocol handles message
acknowledgement, and when applicable, message sequencing. It is better
suited to the specific task of moving I/O data than the more general
TCP facilities. SIXNET Universal Protocol deals with a lost message retry
or an extra reply more efficiently than the TCP stack, minimizing the
delays in I/O updating caused by a lost message and allowing better control
of retry timeouts by the program. SIXNET communications are field proven
over inherently unreliable media such as radio links and telephone lines.
In the case of direct connections via a single Ethernet LAN, there is high
reliability, and these issues shouldn't be significant to the performance
of your system.