Sun-Spots Digest, v6n21

[William LeFebvre]

SUN-SPOTS DIGEST        Thursday, 25 February 1988     Volume 6 : Issue 21

Today's Topics:
         Re: Setting up diskless clients without using Setup (2)
                   Re: Mapping bad blocks to file names
                        Re: Overlapping partitions
                       Re: Two monitors on one Sun
      Re: lpd printing to a TCP terminal server attached printer (2)
                              Re: calentool
                            Re: vt100 emulator
                     a few bugs found in SunOS 4-3.2
                             uucp throughput
                                 SCSI bus
                      3/50 shoebox DC power ratings?
                          Disk Setup on Server?
          Where to put swap partition for clients running CLisp?

Send contributions to:  sun-spots at rice.edu
Send subscription add/delete requests to:  sun-spots-request at rice.edu
Bitnet readers can subscribe directly with the CMS command:
    TELL LISTSERV AT RICE SUBSCRIBE SUNSPOTS My Full Name
Recent backissues are stored on "titan.rice.edu".  For volume X, issue Y,
"get sun-spots/vXnY".  They are also accessible through the archive
server:  mail the word "help" to "archive-server at rice.edu".

----------------------------------------------------------------------

Date:    Tue, 16 Feb 88 16:06:17 PST
From:    rusty%velveeta.Berkeley.EDU at berkeley.edu (rusty wright)
Subject: Re: Setting up diskless clients without using Setup (1)

> From: Steve Schlaifer x4-3171 <steve at mahendo.jpl.nasa.gov>
> The hard part was figuring out what goes in the partitions and how to go
> about providing /tmp, /usr/spool/... etc. directories for the client.

I just use dump piped to restore; for example I typically used ndl0 to
make up partitions for new clients:

	% client_mkfs /dev/rndl9 (a shell script with the right mkfs nos.)
	% mount -v /dev/ndl9 /clients/ndl9
	% cd /clients/ndl9
	% dump 0f - /dev/rndl0 | restore rf -
	% umount -v /dev/ndl9
	% fsck /dev/rndl9
	% mount -v /dev/ndl9 /clients/ndl9

Now fix /etc/rc.boot, clean out /usr/adm, etc.  I also avoid a lot of
configuration malarkey by maximizing the number of symbolic links; for
example, /etc/{printcap,gettytab,resolv.conf} /.rhosts, /.cshrc, etc.  are
all symbolic links to files in /usr/local/{etc,root,...}.  For the stuff
that has to be on the client's root I have master copies in /usr/local and
just cp -p them to the client.

For me the hard part is figuring out what numbers to put in /etc/nd.local;
arithmetic was never my forte.

------------------------------

Date:    Tue, 16 Feb 88 20:17:21 EST
From:    cognos!todds at zorac.arpa (Todd Sandor)
Subject: Re: Setting up diskless clients without using Setup (2)

We had to add a new disk drive and four new 3/50's to the configuration of
a 3/160(2 disks), 3/180 (1 disk) and 18 workstations(3/50).  Were adding
drive to 3/180 and were setting up 10 WS on the 3/160 and 12 on the 3/180.
In our situation we were also expanding the size of our user parti- tions
(extra disk) as well as increasing the swap space size of each of the 3/50
WS from 16M to 20M, thus locations of all partitions except servers root,
swap and the ndl0 root partitions were changing.  Phoned Sun in Calif. and
they said they don't support or recommend trying what I was going to do
but that I should use setup. Yea, right!  We didn't, and everything worked
out fine and finished much faster than if we would have used setup.  Have
outlined basic steps below, have left out some of our details as steps to
take would depend on exactly what has to be done at a particular site.
Just included steps necessary if added new disk-less clients.  Remember to
PLAN things out before hand, and go over plans at least once.

Steps

1) Backup everything, use dump level 0, even dump off a least one of your
current nd partitions as will it be used to reload your nd root partitions
later.

2) Next step is to recreate your new pub partition, this is the trick
which saves the time.  Due to fact that booting up single user requires
mounting of /pub, must have pub all the time.  The trick is to shutdown
system and boot up with stand/diag (>b stand/diag).  Then you re-write a
temperary disk label so can create the new pub partition.  Can do this as
long as new pub will not overlap your old pub, you should be able to plan
it so it doesn't.  Leave size of pub parti- tion same size.  After use
partition and then label commands in diag then halt it and reboot unix up
single user.  This temp. label will contain same root,swap,pub as before,
addi- tion of new pub partition.  During this procedure you will be
destroying the old partition where the new pub is going.  After up single
user you:

# Create new /pub from current one
# newpub is xy0h
/etc/newfs -v /dev/rxy0h
cd /
mkdir pubnew
/etc/mount /dev/xy0h /pubnew
cd /pub
tar cfv - . | (cd /pubnew; tar xfp -)
/etc/umount /pubnew
cd /
rmdir pubnew

At this point shutdown the system and boot stand/diag again, this time
rewriting label so xy0f (pub) describes the new pub, and change xy0h to
zeroes.  To be safe you can re-boot single-user again, to make sure comes
up ok with the new pub partition.

3) Next you try to re-create any partition you can in a similar manner to
above, thus allowing disk to disk parti- tion recreations rather that tape
to disk restores.  In our situtation we were able to recreate /usr and a
user parti- tion.  Again basic step is to re-write the label so both old
and new partitions do not overlap then bring up system single-user then
use similar (obviously different partitions and dir names) commands to
those show above.

4) You re-create all the disk partitions and restore all partitions (via
tape when have to) in single-user mode, no need to go multi-user (and you
couldn't anyway for a majori- ty of the steps).

5) Next thing to do before you bring system up multi-user is to recreate
the nd partitions from the level 0 dump tape that was created before you
started everything.  Create file called ndcreate with following:

---cut-here---------------------------------------
#!/bin/csh
cd /
echo "making /dev/ndl$1 as $2"
# NOTE: this mkfs changes based on root partition size
# and type of disk drive you are using  (PLAN AHEAD)
/etc/mkfs /dev/ndl$1 16560 46 20 8192 1024
/mount /dev/ndl$1 /mnt
cd /mnt
restore rvbf 126 /dev/rmt8
cd /mnt/etc
ed rc.boot <<EOF
/originalhostname/s//$2/
w
q
EOF
cd /
umount /mnt
--cut-here-------------------------------------------

Where originalhostname is name of WS that was used for dumping.  Then
simply create ndall file with

ndcreate 0 host0
ndcreate 1 host1
ndcreate 2 host2


------------------------------

Date:    Mon, 15 Feb 88 19:04:55 PST
From:    parlier at jpl-devvax.jpl.nasa.gov (Randy Parlier)
Subject: Re: Mapping bad blocks to file names

Another way to get the file name would be to use the icheck -b option to
find the inode # and then use 'find' with the '-inum' option.

Randy Parlier

------------------------------

Date:    16 Feb 88 16:48:42 GMT
From:    Jeff Glass <jjg%linus at mitre-bedford.arpa>
Subject: Re: Overlapping partitions

I use the following script to check /etc/nd.local for overlapping
partitions.  I was going to submit it after I had hacked it some more
(have it check for gaps between partitions, for example) but it appears
that now is the time to post it.

It ignores comments and other extraneous lines (e.g., "son").  It expects
one argument, the name of the file to check (usually /etc/nd.local).

It handles nd partitions (e.g., /dev/ndl0 and /dev/ndl1) on more than one
hard partition (e.g., /dev/xy1c and /dev/xy2c).  For each hard partition,
it prints out the nd partitions that overlap.  It does not check if you
have overlapping hard partitions.

nd partitions are printed as "hostname unitnumber" (e.g., "daffy 0" is
daffy's root partition).  At the end, it prints the number of errors
found.

# to unbundle, sh this file.
# files: /guru/jjg/bin/check_nd
echo 'check_nd_gap'
/bin/sed -e 's/^X//' > 'check_nd' << 'End of /guru/jjg/tmp/check_nd'
X#! /bin/csh -fb
X#
X# jjg Tue Feb 16 17:46:36 EST 1988
X#
X# check_nd [-g] [file1 ...]
X#
X# checks files in /etc/nd.local format, looking for nd partitions which
X# overlap, and optionally, which have gaps between them.
X#
X# file1 can be a pathname or "-" for stdin.  If there are no arguments,
X# check_nd reads from stdin.
X#
X# check_nd prints the set of errors it finds, and at the end prints out
X# a summary of the errors.
X#
X
X# look for options.
Xif ( " $1" == " -g" ) then
X	set flag_g = 1
X	shift
Xelse
X	set flag_g = 0
Xendif
X
X# if no arguments, read stdin.
Xif ( $#argv == 0 ) set argv = ( - )
X
Xforeach i ( $argv:q )
X	cat $i:q | \
X	grep '^[ 	]*user[ 	]' | \
X	sort +3 -4 +4n -5 +5n -6 | \
X	awk '\
X	$4 != last_fsys { \
X			  last_part = "the beginning of " $4 \
X			  last_end = 0 \
X			  last_fsys = $4 \
X			} \
X	$6 > 0		{ \
X			  part = $2 " " $3 \
X			  begin = $5 \
X			  end = $5 + $6 \
X			  if ( begin == last_end ) { \
X			  } else if ( begin < last_end ) { \
X			    printf "There is an overlap between \"%s\" and \"%s\".\n", last_part, part \
X			    errs_overlap++ \
X			  } else if ( flag_g > 0 ) { \
X			    printf "There is a gap between \"%s\" and \"%s\".\n", last_part, part \
X			    errs_gap++ \
X			  } \
X			  last_end = end \
X			  last_part = part \
X			} \
X	END		{ \
X			  printf "Errors: " \
X			  if ( errs_overlap == 1 ) \
X			    printf "%d overlap", errs_overlap \
X			  else if ( errs_overlap > 1 ) \
X			    printf "%d overlaps", errs_overlap \
X			  if ( errs_overlap > 0 && errs_gap > 0 ) \
X			    printf ", " \
X			  if ( errs_gap == 1 ) \
X			    printf "%d gap", errs_gap \
X			  else if ( errs_gap > 1 ) \
X			    printf "%d gaps", errs_gap \
X			  printf ".\n" \
X			}' flag_g=$flag_g -
Xend
End of /guru/jjg/tmp/check_nd

[[ It looks very useful, perhaps even indispensable, for nd.local hackers.
Thank you!  I have also archived it under the name
"sun-source/check_nd.shar".  --wnl ]]

------------------------------

Date:    Wed, 17 Feb 88 07:45:07 EST
From:    Chuck Musciano <chuck at trantor.harris-atd.com>
Subject: Re: Two monitors on one Sun

Here at Harris, we had a need to do demos for large groups who could see
the Sun monitor clearly.  They all couldn't crowd around one, so we made a
simple Y connector to hook two heads to one video port on the back of my
3/50.  We used the expensive, fully shielded 9-wire cable, and the big,
solid metal connectors to provide good ground.  It's a bit of a squeeze
getting two cables into the single connector at the base of the Y, but our
technician did a good job grounding both cables and covered the whole mess
with heat-shrink tubing.  The Y connector worked fine.  The cable was
actually a "V" connector, with each leg about 10 feet long.  We saw no
real image degradation at this length, but putting a 16 foot extender on
one leg caused that monitor to seriously smear.  The demos went well, too,
because the second head did not require an operator in front of it, so
people could get closer to it.

Chuck Musciano
Advanced Technology Department
Harris Corporation
(305) 727-6131
ARPA: chuck at trantor.harris-atd.com

------------------------------

Date:    Wed, 17 Feb 88 08:31 EDT
From:    Simon Barnes +44 223 315576 <BARNES at scrvx2.sdr.slb.com>
Subject: Re: lpd printing to a TCP terminal server attached printer (1)

There are several ways of directing printout to a port on a TCP/IP
terminal server.

1. If a filter already exists for the printer and you have source code for
it, then you can modify it to open a TCP/IP connection to the terminal
server and write to the TCP/IP socket instead of standard output.   

2. If a filter exists but you don't have source code,  you can write an
auxiliary filter which opens the socket, points standard output at the
socket (use dup2()) and then exec's the original filter.

3. If no filter exists,  write one or use approach 2 and exec cat.

Modifying lpd should never be necessary, although one hopes that Sun will
get around to building in TCP/IP support some time to save everybody from
having to re-invent these hacks (however entertaining they may be).

Simon Barnes
Cambridge UK

------------------------------

Date:    Wed, 17 Feb 88 11:17:46 EST
From:    marantz at aramis.rutgers.edu (Roy Marantz)
Subject: Re: lpd printing to a TCP terminal server attached printer (2)

Eriks Rugelis asks the magical questions:

>   1. How do I get lpd to speak to a raw TCP datastream?
>   2. Do I have to make source mods to lpd...?
>   3. Does anyone else out there have some tools/sources/hints that they
>      could throw our way to make this job a little easier?

We are finishing debugging a hacked version of telnet which we'll use to
"pipe" between a TCP stream and a unix "device".  We need to use this
method in order to have a full duplex "device" so a postscript printer at
the far end will work.  We haven't modified lpd nor any transcript
software.  I can't give you the source (unless you've got a source
license), but you could write such a beast fairly easily from scratch.  

Roy
--
uucp:   {ames, cbosgd, harvard, moss}!aramis.rutgers.edu!marantz
arpa:   marantz at aramis.rutger.edu

------------------------------

Date:    17 Feb 88 13:36:46 GMT
From:    Sarah Metcalfe <sarahm at cognos.uucp>
Subject: Re: calentool

I wrote a program called Monthtool which may be what you are looking for.
It was designed for monochrome Sun 3/50's.  It was posted to the newsgroup
comp.sources.unix and would have been archived.  It is also available from
me directly -- perhaps preferably as I have a new version that is quite an
improvement over the old.  I'm not sure what calentool does, though, so I
can't say if Monthtool is a valid alternative.  Judging from the mail I
get, it sounds like Monthtool is fairly widely used now, and some people
swear by it (plug, plug).

Sarah Metcalfe         decvax!utzoo!dciem!nrcaer!cognos!sarahm
Cognos Incorporated    P.O. Box 9707, 3755 Riverside Drive, 
                       Ottawa, Ontario, CANADA  K1G 3Z4
                       (613) 738-1440

------------------------------

Date:    Wed, 17 Feb 88 10:13:18 EST
From:    ted at braggvax.arpa
Subject: Re: vt100 emulator

About a vt100 emulator, Edward L. Lafferty of mitre (ell at mitre.arpa) has a
vt100 tool that includes keypad support, line drawing and other special
characters (pi, British Pound, diamonds etc), 132 col mode, double width
and height characters.  I know there are versions for sun 2.X and 3.X.  If
you can't get with Mr. Lafferty (who I found quite helpful the time I
corresponed with him) for the latest version, I guess I can put a binary
up for ftp from braggvax.

Ted Nolan
ted at braggvax.arpa

------------------------------

Date:    Wed, 17 Feb 88 00:15:35 EST
From:    schwartz at gondor.cs.psu.edu (Scott E. Schwartz)
Subject: a few bugs found in SunOS 4-3.2

Sun4 users, here are a couple of bugs I've noticed recently in 4-3.2.
First, if you attempt to use the system V version of curses, and your
terminal is not in the terminfo database, curses will die with a core dump
when you do initscr().

Second, nroff and troff fail in strange ways.  Try:
	troff -rc1 /dev/null
and notice that troff wants to read stdin.  Try:
	nroff -rc1 /dev/null
and notice that it dump core when you give stdin a ctrl-d.

Sun has been notified of both these bugs.

-- Scott Schwartz
	schwartz at psuvax1.cs.psu.edu
                     ^
                     | Someday soon this will be a sun4, but the name will
                       stay the same! :-)

------------------------------

Date:    Tue, 16 Feb 88 10:45:47 PST
From:    coherent!dplatt at ames.arc.nasa.gov (Dave Platt)
Subject: uucp throughput

For what it's worth:  on our Sun 3/280, it seems to take about 60% of the
CPU to suck input from a Telebit TrailBlazer Plus, with the serial port
speed set to 19200 and overall throughput about 12000 baud (e.g. 1.2k
bytes/second).  If we try to glom input from the 'Blazer and from a
9600-baud direct-connected neighbor (aimt), we tend to peg the CPU meter.

It doesn't seem to matter much whether the 'Blazer and hardwired line are
connected to the CPU-board serial ports or to the ALM-2... the CPU load
didn't seem to change much when we installed the ALM-2 and moved the modem
links over to there.

I haven't looked at perfmon to see how much of the load is user-state and
how much is system-level.

Dave Platt
  UUCP:	...!{ames,sun,uunet}!coherent!dplatt     Domain: dplatt at coherent.com
      Internet: coherent!dplatt at ames.arpa, ... at sun.com, ... at uunet.uu.net

------------------------------

Date:    Tue, 16 Feb 88 14:26:56 PST
From:    malik at sun.com (Malik Shaikh)
Subject: SCSI bus

Learn more about scsi bus signals, espacialy four control signals, these
signals put SCSI bus in one of eight phases

        BUS Free phase
        ARBITRATION phase
        SELECTION phase
        COMMAND phase    \
        DATA phase        \  These phases are collectivelly termed
        STATUS phase      /  the information transfer phases
        MESSAGE phase    /

During building system I got message that SCSI select failed so I realized
that HBA have problem in selecting device so I realized that scsi HBA is
not good.

------------------------------

Date:    Tue, 23 Feb 88 20:05:28 PST
From:    merat at uw-ftcl.ee.washington.edu (Merat Bagha)
Subject: 3/50 shoebox DC power ratings?

Do you know the DC power ratings of a 3/50 shoebox power supply?

Specifically, I would like to know if I can attach an external disk drive
which draws power from the P14 power connector which is left unattached
inside the shoebox.  

My Fujitsu M2246E (141MB ESDI) requires:

    1.8 Amps @ 12V & 1.6 Amps @ 5V (average);
    4.8 Amps @ 12V (peak).

Many thanks in advance!

Merat Bagha
Univ. of Washington, Seattle

------------------------------

Date:    Tue Feb 16 18:27:06 1988
From:    portal!cup.portal.com!mhyman at sun.com
Subject: Disk Setup on Server?

Where I work we are installing a ``pilot'' network consisting of a 3/280
server with two 580 Meg drives and 6 diskless 6/60's with 16 Mbytes of RAM
each.

All of us here are Sun neophytes.  Our question is how would real live
users recommend partitioning the disk?  Is the default root size ok for
the server?  for the clients?  How much swap space should we reserve for
the server?  For the clients?

We would appreciate any and all advice.  Please mail to:

   mhyman at cup.portal.com  --or--
   ...!sun!portal!cup.portal.com!mhyman

Thanks,
--Marc

------------------------------

Date:    Wed, 17 Feb 88 14:40:03 est
From:    petsche at demon.siemens.com (Tom Petsche)
Subject: Where to put swap partition for clients running CLisp?

There are three people here interested in running Common Lisp on Sun
3/60s.  We currently have a server with two Fujitsu M2344k SMD drives and
three clients.  The question is what should we do for swap devices given
that Lisp is a memory hog. 

If the 3/60s swap to the server, will this severly degrade performance on
the clients?  More importantly, will this degrade the network or server
performance?  In short, does it make more sense to put the swap space on
the server or get small (70-100 Meg) local SCSI disks for the machines
running Lisp and put the swap and root partitions on these drives?

Thanks.

tom  (petsche at siemens.com)

------------------------------

End of SUN-Spots Digest
***********************