Some Internal Tymnet Protocols
The virtues of various Internet protocols have been well described in various publications.
Here is a fragmentary sketch of some Tymnet protocols and advantages some of them relative to those of Internet.
Tymnet Link Protocols
The basic contract of the network proper is to carry streams (not packets) of characters.
A circuit is full duplex and follows a route fixed upon creation (login) by a central network supervisor.
A circuit is composed of two streams, coming and going.
There are just a few escape characters that are significant to the intermediate nodes and a few more significant to the edge node that communicates directly with the ASCII terminal.
The same stream format is delivered to and accepted from the various Tymnet-native hosts.
Tymnet itself imposes no packet format on the streams.
Information on a link between two Tymnet nodes is organized into packets and error controlled separately on each link.
A link transmission error is repaired with no extra traffic except on that link.
Hosts are not involved in this error control.
Each node has a byte queue for each stream passing thru the node.
Packets for a link are composed of bytes from active streams.
They are assembled just before transmission from queues with characters for that link.
Flow control, which we called “back-pressure” is also link by link and stream by stream.
Twice a second a control packet flows upstream on each link with a bit per downstream stream.
This provides “bang-bang” control.
Lost control packets were taken to mean “stop”.
Laroy says more about these things.
Routing in Tymnet
At each Tymnet node there is for each circuit thru that node:
- A pair of variable size character buffers, one for each direction,
- Identification of the two links on which the circuit passed the node,
- For both of these links an 8 bit circuit identifier by which the node at the other end of the link would identify traffic for the circuit.
- A bandwidth indicator.
(Originally there were 0 bits, then 1.
Later there may have been more.)
For each link at a node there is:
- a table of buffer pair locators.
The table is indexed by the 8 bit circuit names from traffic arriving on the link.
- A parallel table of bits indicating a downstream invitation for more data.
Physical packets are check-summed and classic error control is performed between nodes.
A packet carries many segments which are each data from a different circuit.
Each segment carries its 8 bit circuit ID and character count.
Note that these 8 bit circuit IDs are typically different for each link that a circuit traverses.
They are like Cisco’s tag in their “tag routing” scheme, or perhaps like the labels in the MPLS proposal.
Efficiency, vs. Internet
There are several efficiencies in these schemes.
Note that the traffic on a link has better thru-put when the link is backlogged.
That is because the physical “packets” on the link are larger under load, and have a favorable ratio of payload to header.
Under lighter load they are small which provides better latency.
Due to link by link flow control there are virtually no lost packets.
Internet packets lost on one link (due to congestion) may cause extra traffic on other links (due to end-to-end retransmission).
This phenomenon can lead to gridlock.
The “slow start” protocol, presumably governing most Internet traffic, is unable to employ sporadic dips in link traffic.
In Tymnet, by contrast, there are many bytes queued for any congested link ready to go on a millisecond’s notice, should the load decrease.
Single keystrokes, now less common perhaps, travel thru the congested net using 24 bits on each link including overhead.
This is the portion of a physical packet consumed by an isolated character on a circuit.