home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!shady!kevin
- From: kevin@shady.UUCP (Kevin Smith)
- Newsgroups: biz.sco.general
- Subject: Re: time command
- Message-ID: <77@shady.UUCP>
- Date: 31 Dec 92 20:54:56 GMT
- References: <C03p7q.9v1@world.std.com> <1992Dec31.043015.28653@netcom.com>
- Organization: ShadeTree Software, Inc.
- Lines: 58
-
- In article <1992Dec31.043015.28653@netcom.com> gerg@netcom.com (Greg Andrews) writes:
- :>apl@world.std.com (Anthony P Lawrence) writes:
- :>>
- :>>I'm sitting here playing with a 115K baud terminal. I wanted to
- :>>see how it compares to the system console, so I did 'time lr /etc'
- :>>on both the console and the terminal.
- :>>
- :>>The console finishes first, as we would all expect. But, 'time'
- :>>reports LESS real, user and system time for the command run on
- :>>the terminal.
- :>>
- :>>The only rational explanation is that either the multiport card or the
- :>>terminal (or both) are doing a lot of buffering so that 'time' thinks
- :>>the command is complete when it actually isn't.
- :>>
- :>
- :>I see it differently. Running a 386 box from a terminal is a primitive
- :>form of "client/server" computing. Running it from the console is not.
- :>
- :>With a terminal, your application simply shoves its output to the
- :>terminal. The terminal does the work of placing the character into
- :>the screen buffer, moving the cursor to the next position (which can
- :>include wrapping to the first column of the next line), and scrolling
- :>the screen if you're on the last line.
- :>
- :>When you're on the console, the computer must handle your command *and*
- :>perform all the character and cursor manipulations for the screen. Thus,
- :>the work that normally would be off-loaded to the terminal is seen by the
- :>cpu and it leaves that much less available to the rest of the system.
- :>
- :>--
-
- As I see it, real time is real time. Since the real time to the
- terminal was shorter than the console or the "observed" time the only
- answer is buffering in the serial controller or driver or the
- terminal. The user and system times reflect how much effort the cpu
- put into the io. Interrupt driven dumb serial controllers can kill a
- cpu at 9600 baud. Intellegent controllers can impose a wide range of
- loads depending just about everything. A recent issue of SCO magazine
- had a multiport serial controller review that addressed actual CPU
- loads from different boards. Also keep in mind that sending data out
- a port is always easier that receiving. The computational load of
- managing the console screen is probably pretty small.
-
- You should put a stop watch on it and figure the actual transfer rate
- to the screen. Furthermore you should cat a large file rather than
- 'lr' since listing files involves far more cpu and disk processing.
- FurtherFurthermore a lot of terminals cannot actually handle continous
- input at their rated speed and will revert to some form of handshaking
- to moderate their input. I like to create a huge file of lines of 79
- '#'s. Disable any terminal handshaking and let it fly. Watch the
- output for ragged right margins that indicate data loss. This is
- also good for testing printers.
- --
- | Email - !shady!kevin uunet!shady!kevin kevin%shady@uunet.uu.net
- Kevin Smith | Voice - (+1) (908) 874-7980
- | Mail - ShadeTree Software, Inc., 192 Capricorn Dr. #10,
- | Somerville, NJ 08876, USA
-