Chapter 26

p. 392

The recommendation against using .rhosts files is certainly as valid now as then. Especially in Universities (and Bath is no exception), a Kerberos-based system is used instead.

p. 395

The comparison between client$\rightarrow$server signalling and server$\rightarrow$client signalling is not quite as clear (in my opinion) as Stevens makes out. In particular, Telnet uses (essentially)⁵² in-band in both directions, but has a true escape mechanism, whereas rlogin does not have a way of sending the escape character 0xff as data.

p. 398

Logically (especially in the context of large fat networks), step 8 need only follow step 2, but in a LAN context (where rlogin is normally used) it will follow step 7.

pp. 401-3

telnet is fundamentally trying to solve a $N^2$ problem: if there are $N$ different kinds of terminal, there are $N^2$ possible mappings. Mapping to/from NVT makes it a $2N$ problem, but at the cost of crippling some functionality/response, so a compromise has to be sought. This explains the graph of code size in Figure 26.1

Figure 26.13

Segments 7 and 8 have, due to the delayed ACK rule, a 20% chance of being combined. So an option can be acknowledged at the same time as it is being replied to, but the TCP client cannot see this.

p. 413

Of course, if the Nagle algorithm is disabled, then every character typed is sent in a separate segment.