3.51m kernel and uninterruptible power supplies
Bill Mayhew
wtm at uhura.neoucom.EDU
Sat Mar 10 22:35:46 AEST 1990
My 3b1 has been on a standby power system since late in 1987. The
transfer timeof my standby system is about 7 mS, so it doesn't
qualify as truly uniterruptible. I do believe the 3b1 power pack
has enough ride-through to handle that much of a gap, though.
Despite being properly grounded and powered, my 3.51m did lock up
once while doing a cpio operation. Looking at /usr/adm/unix.log
didn't show any error messges and a subsequent surface scan of the
hard disk didn't reveal any bad sectors. Cosmic rays maybe?...
The kernel panic printed over top of what was already on the
screen, including the meter maid display, so it was rather
difficult to get any information from the dump.
Another thing I noticed about 3.51m is that it takes 3.51 longer to
notice an in-band flow control from the tty000 port. I believe I
mentioned that here before. I use a trailblazer plus, but have to
support dial-ins at various baud rates (especially myself from my
portable peecee at 2400 baud) as well as PEP mode for uucp. Yes
Mom, I use hardware flow control from the tb+ to the computer.
That works great when the tb+ decides things need to slow down.
The problem is when a remote sloth like my peecee clone portable
wants to do an in-band flow control. Obviously the modem can't do
something brash like convert the xoff to an out-of-band hardware
flow control, or binary transfers would be seriously mangled.
Now the moral of the story is that things worked correctly with the
3.51 kernel, but 3.51m loses characters when the remote does
in-band flow control. Perhaps the incoming queue is longer with
3.51m, thus increasing latency of servicing in-band flow
control?... If so, is the queue length adjustible? Apologies if
my question is dumb, but I am not an inveterate kernel hacker.
Bill
--
Bill Mayhew Northeastern Ohio Universities College of Medicine
Rootstown, OH 44272-9995 USA 216-325-2511
wtm at uhura.neoucom.edu ....!uunet!aablue!neoucom!wtm
More information about the Comp.sys.att
mailing list