TTY performance measurement

[John F. Haugh II]

In article <138 at wobble.UUCP> dlu at wobble.UUCP (Doug Urner) writes:
>I'm looking for information on measuring tty sub-system performance.
>Pointers to articles and/or benchmarks, anecdotes and insight into the
>issues involved would all be appreciated.

Once I was charged with testing the I/O performance on a certain
vendor's 68000 system.  The tty sub-system was controlled by a 12MHz
68000 running without interrupts polling a collection of dual UARTS.

My test was to copy a single character from one port to the next, and
back, counting the number of characters per second.  The ports were set
for 38.4KBaud, so I expected several hundred characters per second ...

The result was < 50 cps.  Turns out the polling was driven by a 10ms
clock.  Every 10ms all the UARTS were checked for incoming characters,
so the process would invariably sleep for at least 10ms waiting on the
character it had just now been sent by the producing process.  Or
something like that.

Moral:  When testing tty systems, always try at least one perverse
case.
-- 
John F. Haugh II                        +----Make believe quote of the week----
VoiceNet: (214) 250-3311   Data: -6272  | Nancy Reagan on Artifical Trish:
InterNet: jfh at rpp386.Dallas.TX.US       |      "Just say `No, Honey'"
UucpNet : <backbone>!killer!rpp386!jfh  +--------------------------------------