Microport: how good

M.R.Murphy mrm at sceard.Sceard.COM
Fri Dec 7 13:07:30 AEST 1990


In article <37 at opel.COM> johnk at opel.COM (John Kennedy) writes:
>In article <1990Nov30.140527.15142 at sbcs.sunysb.edu> pmartino at csserv1.ic.sunysb.edu (Steve Wechsler) writes:
>>
>>Greetins, folks.  I just got hold of a copy of Microport SYS V/AT v2.4 [...]
>[...]
>It's um, pretty good, all things considered.  However, some gotchas:
>
>
>Their fsck is broken, for large file systems, and I can't remember how
>big "large" is.  fsck will actually leave the file system more corrupted
>in some cases.  The workaround is with a replacement fsck binary that was
>floating around the net, that someone sent me, and have around here somewhere.
>
>The biggest gotcha is not so much Microport's fault, but lies with the segmented
>architecture of the 286 itself.  Much of the public domain software doesn't need
>to worry about segment sizes or large and small models of the Intel
>architecture.  Most stuff compiles ok, or has options for 286s.  Other software
>needs to be examined, usually after an attempt at compiling fails.
>
>- Among the software that worked ok:
>
>  News, smail2.5, fas async driver 
>  Any device driver that you want to write can be installed ok.
>
>
>- Among what wouldn't work for me (others may have had better success)
>
>  GNU make (got compiler "register errors")
>
>  Don't expect networking, X windows, or anything that expects STREAMS
>   capabilities, since it is SvsV release 2, not 3.
>
>  Even though I bought it from Microport and tried on three different
>  motherboards with different BIOSs from the approved list, I could never
>  get Dos-Merge to work.  Could have been pilot error, though.
>
>  Although it should run on a 386 board, I tried it and got erratic keyboard
>  operation.  No explanation.
>
>This sounds like a good project if you have plenty of time to spare and
>treat it is a Unix learning experience.  If you're in a hurry to get
>a system up and stable, there are easier ways.  If you're looking
>for a System V to dig into and try to understand, you've got it.
>
>John
>
>
>
>-- 
>John Kennedy                     johnk at opel.COM
>Second Source, Inc.
>Annapolis, MD

I've been running uPort 2.4 for quite some time now. The system is stable
enough so that I notice the feature (bug:-) in ruptime that uptimes over
90 days are reported as question marks. The key things to do to make this
sort of system work are:

  1) get the inode fix from Bob Thrush (rd at aii.com, uunet!tarpit!rd),
  2) get the fixed fsck from John Kennedy :-),
  3) put in the fas tty driver or the other really good driver from Jim Murray,
  4) use the 2.3 console stuff,
  5) re-link the system and increase number of process, clists, and a bunch
     of other things. Fortunately all of the parameters are completely described
     in the documentation (dreamer),
  6) use smail2.5 and deliver or smail3.1,
  7) use HDB UUCP,
  8) use C News in small model (B News ran fine, only in large model, but the
     system ran flat out of cpu time as in 0% idle time for four days with an
     ever-growing backlog of news. After switching to C News, the system runs
     with about 50% idle time :-)

I've run 2.4 with Excelan TCP/IP (EXOS 205T or EXOS 205E), and with
Whomever-Interlan TCP/IP (NP600). Both are OK, not as good as the latest UCB
stuff, but OK. I've run it on a real 6MHz AT with WD1003/ST251 on through a
16MHz 0ws clone with IDE. Faster and bigger is better :-). 2.4 is generally
livelier on a 6386 (4MB, ESDI) than is AT&T SV3.2. VM Unix(tm) is not all good.
Small is sometimes a whole lot better, depending...

Raw disk throughput varies from 100KB/sec (WD1003/ST251) to 700KB/sec
(IDE/Connor 200MB). I haven't had a cache disk controller to try, but I'd like
to hear from someone who has tried one.

2.4 is stable in all hardware configurations I've tried after making the fixes
in 1-5) above. Some disk controllers don't work for beans, sell 'em to somebody
who only needs MS or PC DOS. 16550A chips make unintelligent serial boards work
quite well with the Murray driver, four TB+ at 19.2K locked speed and no lost
characters using RTS/CTS. 2.4 also works well with Overland Data 9tr magtape
systems. It reads V7 tar tapes just fine. Also reads 20 year old tapes that
were written on a 360/75. Luck. Reads and writes tapes to interchange data
with a VAX(tm) 750 and 4.3 BSD, too. Remote backup (both systems running 2.4)
using a command like
  (cd /usr/stuff;find . -print|cpio -oc)|rsh otherhost dd of=/dev/mt/0mn bs=10k
works. So does the restore :-) It also works if one of the systems is 2.4 and
the other is some little grey generic Unix(tm) box with a big monitor.

All in all, uPort 2.4 is as John Kennedy says, pretty good, considering, with
a few gotchas.

Now for a good story. I just had a disk cough with a bad sector
(cyl875,head5,sec1).  I didn't want to backup and restore everything, and since
875,5,1 was in the News spool partition, and therefore couldn't contain anything
of value, I decided to RTFM and see what tools were there to fix things.
Fortunately bdblk(1) - print, initialize, update or recover bad sector
information on disk packs (/etc/bdblk option unit [sector...]) is well
documented in the System V/AT Runtime System manual. Unfortunately, it is not
supplied...  thanks uPort :-) However, fdisk can be used to recover without
clobbering everything. The way to do it is:

   1) backup everything you can,
   2) make sure that initdefault is single user,
   3) sync a few times,
   4) invoke fdisk,
   5) advance disk to next if required,
   6) select 5 - bad track mapping,
   7) don't scan the disk. DON'T SCAN THE DISK, respond n to the question,
   8) don't write a new bad track table, respond n to the question,
   9) enter the additional bad track information, end with ctrl-d
  10) exit back to the main fdisk menu,
  11) be prepared for the reboot when you exit fdisk (respond y to the update
      question),
  12) fsck after the reboot,
  13) use showbad to see what you've done and be amazed that it worked,
  14) change initdefault to 2 if you're daring.

This worked for me. Your mileage may vary.

BTW, uPort, where are send and gath? What if I really had the hots to run
RJE? And where is /etc/bdblk?

-- 
Mike Murphy  mrm at Sceard.COM  ucsd!sceard!mrm  +1 619 598 5874



More information about the Comp.unix.sysv286 mailing list