Welcome to comp.unix.sysv386 (AKA Frequently Asked Questions)

Conor P. Cahill cpcahil at virtech.uucp
Mon May 6 12:47:13 AEST 1991


Welcome to the comp.unix.sysv386 newsgroup.  This newsgroup (actually the
former comp.unix.i386 newsgroup which was renamed to this group) was founded 
to perpetuate discussions of the various unix systems running on the
Intel 80386 architecture.

This article contains the answers to some Frequently Asked Questions
often seen in comp.unix.sysv386. 

Before posting a question to this newsgroup, PLEASE read through this
posting, the Frequently Asked Questions posting in comp.unix.questions, 
and finally the various postings in news.announce.newusers. 

	Last Modified: $Id: freq.ques,v 1.9 91/05/05 22:42:42 cpcahil Exp $

This article includes answers to:

	 1. What is this group for?
	 2. Which Unix should I buy?
	 3. Is there a BSD port for the 386 available anywhere?
	 4. What hardware works with brand X Unix and/or X11
	 5. How do I rewind or retension a cartridge tape?
	 6. How do I get a second serial port working?
	 7. How do I get fast serial line throughput with stock asynch ports?
	 8. Is there a port of X11 R3/4 available for whatever?
	 9. What are good intelligent multi-port serial card solutions?
	10. Where do I get troff from?
	11. Why doesn't vi work on large files?
	12. Why can't I create a file larger than 2MB (was: How do I
	    increase my ulimit)?
	13. How do you set up a port for both dial-in and dial-out access?
	14. How do I setup a global environment variable that even gets set
	    in the /etc/rc*/* scripts.
	15. How do I get more than 8 virtual terminals to work on the console?
	16. Why do I get the error "ps: unlink() error" every once in a while?
	17. How do I set up more than 4 partitions on the disk drive
	18. Why do I run out of inodes on some filesystems when I know I don't
	    have that many files?
	19. What is the minimum hardware for running Unix on a 386?
	20. Will a caching controller be a good investment?
	21. How many users can I run on my 386?
	22. What is the bandwidth of an AT Bus?
	23. Can Unix make use of memory in 16 bit AT bus slots?
	24. How do I fix the error "Out of DOS Inodes" ?
	25. Where can I get the K-Shell (aka ksh)?
	26. What dos-under-unix product will work with ESIX?
	27. How do I correctly configure the various STREAMS parameters?


Before I start on the answers, I will state that there is NO GUARANTEE as 
to the correctness and/or appropriateness of these answers.  If you intend
to try one, BE SURE TO HAVE A BACKUP of your system first.

My current system is Interactive's ISC UNIX 2.2.  I have tested 
most of the responses on this system.  If you find that another system 
requires a different response, please send me mail explaining what is 
required on your system (you might also want to tell me the OS and version).


1. What is this group for?

	This group was formed to discuss the various implementations of UNIX
	on the Intel 386 architecture.  This includes SCO Unix, 386/ix,
	ISC UNIX, ESIX, Dell, Bell Tech (now Intel) Unix, Microport Unix,
	UHC Unix and, of course, AT&T and any others that I have forgotten.
	
	Since no vendor intends to make a 486 specific version of their OS,
	the use of the various UNIX products on the i486 will also be
	discussed.

2. Which Unix should I buy?
	
	This is a hard question.  Few people have access to all of the
	available Unix implementations and without that access it is 
	hard to be able to recommend any one over the other.  You should
	consider the following factors in making your decision:

		1. price
		2. compatibility with existing software
		3. compatibility with other operating systems
		4. ease of use/control (administration menus)
		5. performance

	You must decide which of those factors is more important to you.

	Most, if not all, application packages developed for one UNIX will
	run on one of the others.  If you find one that does not, blast the
	developers because there is no reason for them to use any vendor
	specific code in their applications (and, IMHO it was a real stupid
	decision on their part).

	Contact information for the various OS manufacturers:

	PRODUCT			VENDOR			Phone contact (sales)
	----------------------  ----------------------  ---------------------
	AT&T UNIX		AT&T			???
	DELL UNIX (SVR3&4)	DELL Computer Corp	800-426-5150
	ESIX			Everex			415-683-2068
	ISC UNIX (SVR3)		Interactive Systems	???
	Mach 			Mt Xinu			???
	Microport UNIX (SVR3&4)	Microport		???
	SCO UNIX & Xenix	Santa Cruz Operation	800-726-8649		
	UHC UNIX (SVR4)		UHC			713-782-2700
	

3. Is there a binary BSD port for the 386 available anywhere?

	No.  However, SVR4 has many BSDisms including symbolic links, job
	control, BSD file system, sockets (implemented  on top of streams).
	It will also contain the SunOS memory mapped files, the Korn shell,
	and many other nifty things.

	SVR4 is currently shipping from Microport, UHC, and DELL.  Intell
	had been shipping SVR4, but has handed its UNIX marketing efforts
	over to ISC.  ISC has announced that it will begin shipping SVR4
	in April.  SCO has no plans to ship a SVR4 product, but "will include
	SVR4 features in its SVR3.2 product."

	One thing that should be noted here is that the System V R4 release
	has MAJOR changes over the R3 releases and won't be stable for a while.
	If you want a stable system, I would suggest that you continue to use
	SVR3.2 until SVR4 stablizes (another 6mos to a year).

	BSD 4.4 will support the 386 architecture as a base system.  This
	means that vendors will have a base BSD system that could be used
	to make a BSD binary release for this architecture.  There are
	several reasons why this probably won't occur:

		1. SVR4 will be enough BSD to satisfy most people

		2. BSD kernel would have to be modified to support the 
		   386 ABI or else it would be closed out of a major 
		   portion of the 386 market.

		3. This is a brand new port and it can be expect that
		   there will be significant problems with the stability
		   of any new product.  This puts it around 2 or 3 years
		   behind the other 386 Unixs in this respect.

	Of course, this does not mean that you will never see BSD4.4 on
	a PC (since you will probably see it if you are somehow connected
	with a university that gets BSD source).  It just means that there
	probably wont be a commercial binary release available to the
	general public.

	The other possibility is MACH.  MACH is 4.3BSD compatible and has
	the 386 as one of its target environments, but since it includes
	much of the BSD kernel, it still requires a standard BSD licensing
	(which is actually a requirement for an AT&T version 7 license).

	At least one company (Mt Xinu) has announced that they will support
	(and distribute) the MACH source code (assuming you have the
	appropriate license) AND has announced the availability of a binary
	product based upon the same.  Support for various PC devices is
	somewhat limited and they (Mt Xinu) see this as a developers 
	product, not an end-user product.
		  
4. What hardware works with brand X Unix and/or X11

	The correct answer to this is a recommendation to call the
	distributor of the X11 that you intend to buy.  The list of
	devices changes all the time and attempting to maintain a list
	here would be ludicrous.

5. How do I rewind or retension a cartridge tape?

	This depends upon the OS and tape device.  The standard at&t 
	distribution includes the tapecntl program (it is even documented
	in the 386/ix manuals), but most of the 386 unix products do not
	provide it.  SCO UNIX has the "tape" program.

	The next place to look is the tape driver manufacturer.  Archive
	does provide a "ctape" program which can perform this function.
	
	If your OS manufacturer doesn't provided it and your tape
	manufacturer doesn't provide it, you need to home grow it yourself.

6. How do I get a second serial port working?

	SCO Unix and Xenix come pre-configured to work with two serial port
	devices.  AT&T Unix auto-configures the number of serial ports.

	The rest of the Unix offerings do not and must be re-configured to
	turn on the second serial port.

	The asy file in /etc/conf/sdevice.d contains the kernel configuration
	information for the standard asynchronous ports.  The file should 
	look something like:

		asy	Y	1	7	1 ....
		asy	N	1	7	1 ....

	This shows that the first asy is enabled, while the second is 
	disabled (the N in column 2).  Change this file to:

		asy	Y	2	7	1 ....
		asy	Y	2	7	1 ....

	and rebuild the kernel using kconfig or /etc/conf/bin/idbuild as
	appropriate for your system.  Be sure to check the interrupt number
	and i/o port addresses to ensure they are correct and don't conflict
	with other devices.  See sdevice(4) for more info.

	Note that the change from a 1 to a 2 in column 3 enables the modem
	control ports which can be used for uugetty.

7. How do I get fast serial line throughput with stock asynch ports?

	The stock serial ports usually will drop characters when run at
	high speeds with incomming data.  The best way to fix this is to
	add the "FAS" driver to your system.  (FAS stands for Final 
	Asynch Solution).  The current version of this driver is 2.08.

	The Readme for the FAS driver suggests that you also replace the
	16450 uart chips on your asynch card with 16550s (I think the 
	cost runs around $20).  The 16550s provide a 16 byte FIFO which
	allows operation at high speeds without loosing characters.

	SCO Unix, ESIX, and ISC UNIX are all reputed to support the 16550
	in FIFO mode, so you probably can make this change without having
	to change the drivers.  See your OS documentation for more info.

8. Is there a port of X11 R3/4 available for whatever?

	X11R3 is available from all of the 386 UNIX distributors.

	X11R4 is *usually* available from the System VR4 vendors.

	X11R4 is also available as a third party product from MetroLink.
	MetroLink's port has support for several high performance cards in
	addition to the standard range of VGA cards. 

	X11R4, as ported by Thomas Roell, is freely redistributable and
	supports a wide variety of VGA boards.  All you need is the X11R4
	distribution from MIT, Roell's patches and voila you get X11R4.  For
	sources to the X11R4 sources you can see the FAQ posting in
	comp.windows.x.

	I have heard very good reports on the performance of this server
	from the many people that have used it, so if you need X or 
	want X11R4, you should check this out.

	This stuff is available from the following sites:
	
		ftp:	flop.informatik.tu-muenchen.de in /pub/i386/X11R4

		uucp:   speed: TB+/PEP phone: 512-346-2339
			system: bigtex
			login: nuucp (no password)
			files: /usr3/X11r4/i386.server/README
			       /usr3/X11r4/i386.server/X386.man
			       /usr3/X11r4/i386.server/Xserver.tar.Z

		uucp:	system: zok (connection info available in the 
			FAQ posting to comp.windows.x)

9. What are good intelligent multi-port serial card solutions?

	This is yet another religous issue.  The following cards are 
	available for AT bus systems.  Note that the cards are listed
	in alphabetical order and that this should not be taken as a
	recommendation of any one card over the others.

	Computone
		Computone provides a series of serial port solutions.  I
		have no experience/pricing info on these cards.

	Digiboard 
		Digiboard provides a series of serial port solutions.  I
		have no experience/pricing info on these cards.

	Equinox
		Equinox has several serial port solutions including 12, 24
		and 96 ports in a single slot.  Each board is reputed to be
		capable of running ALL ports at full speed (38400 baud)
		simultaneously with no loss of data.  The 24 port card runs
		around $1200.  Installation is a snap, no jumpers to
		set, no interrupts required for the board.   The board
		can be installed to use memory between 640K and 1 MB for
		communications between the board and the system.
		The board runs fine.  Input/output port sharing is supported.

	Maxpeed
		Maxpeed has 8 and 16 port cards that will run any of the ports
		at speeds up to 38.4K. The 8 port card retails for 795, but is 
		usually available at $500 to $550.  I lost data when 
		receiving data as speeds > 9600 baud.  VPIX did not 
		run correctly on my wyse 60's.  I contacted customer
		support and got the response "it works fine on other
		machines".
		Other than the vpix problem I had this board provides
		good support for 8 terminals/printers at a low price.  

	Specialix
		Specialix provides a series of serial port solutions.  I
		have no experience/pricing info on these cards.


       Consensys
		Consensys has two serial port solutions in their POWERPORTS
		line.  Each POWERPORTS intelligent main card supports up to 
		8 ports and can optionally accept an 8-port daughtercard.
		Each main card takes a 16-bit slot; the daughters don't
		require backplane access, but they do consume a slot cover.
		Each port is supposed to be able to handle 38.4, but there
		can be problems when driving Trailblazers from a daughter
		port.

		They also provide a multiple-session-per-terminal feature.
		You can drive up to 8 independent sessions from each physical
		terminal, toggling between sessions via control keys.
		The only downside is that you must go through a two-minute
		reconfiguration of the ports software if you switch terminal
		types on a port, since the card must load the appropriate 
		termcap/terminfo.  They also provide software for transparent 
		printing to a printer attached to an aux port on one of the
		terminals.

		List prices range from $850 to $2,000 depending upon port
		and board RAM configurations.

10. Where do I get troff from?

	You can but the Documenters Workbench package from your 
	Unix distributor, or a substitute package from a third party 
	venor.  A second option is to obtain GROFF (which is free and 
	available at your local GNU repository).

	The problem with the standard packages is that they do not support
	most of the currently available printers in troff mode.

	The third party packages seem to fill this niche quite well.
	These include:

		Eroff 			$795 (415-964-2200)
		SoftQuad Publishing	$995 (800-387-2777)
		Jetroff			$69 (personal)/$119 (commercial)
		Psroff			$0

	I have only used Eroff and it works quite well.  I have heard good
	words about SoftQuad and Jetroff.  Psroff was posted to
	comp.sources.unix late last year.

11. Why doesn't vi work on large files?

	There are two possible problems.  The first problem may be
	that you are running into a ULIMIT problem.  See the discussion
	under question 12 to figure out how to change/correct it.

	If this is not the probem, you may be running into a bug/feature
	of some VIs.  The problem is that there are compiled in limits
	as to the number of lines in the data file. 
	
	This problem is known to exist in the following Unix packages:

		AT&T System V Rel 3.2
		Bell Technologies System V Rel 3.2 (was ok in 3.1)

12. Why can't I create a file larger than 2MB (was: How do I increase my
    ulimit)?

	For those of you who don't know, the ulimit is the limit in the
	number of blocks that can be written to a file.  This is intended 
	to be a mechanism to stop a run-away process from eating up all the
	disk space available on your system.

	For UNIX versions prior to SVR4:

	Both SCO Unix and SCO Xenix start out with a ulimit of 2097152 which
	means that you probably won't have a problem with the ulimit on
	these operating systems.  However, if you have one of the AT&T
	derived operating systems, this limit will usually default to
	2048 or 4096 and it has a hard limit of 12288.  This can be 
	unacceptable for systems that will have moderate to large database
	files and will therefore need to be increased.

	1. If your desired limit is > 12288(6MB):

		Edit /etc/conf/cf.d/mtune to change the following line:
			ULIMIT        	3072	2048	12288
		to:
			ULIMIT        	3072	2048	xxxxx

		where xxxxx is the limit you desire.

	2. Edit /etc/conf/cf.d/stune to add/change the following line:

		ULIMIT	xxxxx
	
	   where xxxxx is the limit you desire.  Note that this step can
	   be performed in the kernel configuration software (i.e.: kconfig
	   for 386/ix).

	3. Edit /etc/default/login to delete the ULIMIT line.

	4. Rebuild the kernel and reboot.

	For System VR4 and above:

	While the ULIMIT parameter is still present (even in the
	/etc/default/login file), it is not the real limitor.  The 
	reql variables are: SFSZLIM and HFSZLIM.

		SFSZLIM is the default setting for all programs
		HFSZLIM is the maximum setting of SFSZLIM (i.e. the 
			largest value you can raise SFSZLIM to).

	Note that the units for these variables is different from ULIMIT. 
	Instead of blocks, the units is bytes, so to allow 11MB you need 
	to specify something like 11534336.
	
	In versions of SVR4 prior to release 3.0, if the SFSZLIM is 
	increased by a user and that user subsequently runs a setuid program,
	the kernel resets the SFSZLIM for the exec'd program and any of its
	children.  Starting with release 3.0, this will only happen if the 
	SFSZLIM had been lowered.

13. How do you set up a port for both dial-in and dial-out access?

	There are two ways to do this.  Most intelligent boards provide
	two devices for every port: one with and one without modem
	controls.  The modem control ports are used for getty, while
	the corresponding non-modem control lines are used for cu,
	uucp, kermit, etc.  This is the easiest mechanism to configure
	and use (provided the device driver works properly).

	The SCO and ESIX systems and the FAS driver provide modem control
	ports which can be used in the same manner.

	The second solution is to use the uugetty software. uugetty
	is a replacement for getty that uses uucp/cu locks to control
	access to the port.  In order to work properly,  the port must
	have modem control support.

	386/ix and ESIX have support for the modem control on the serial
	ports, but does not configure these devices automatically.  You
	must mannually create the ports using the following commands:

	For 386/ix:

		mknod /dev/ttym0 c 3 16
		mknod /dev/ttym1 c 3 17

	For ESIX:
	
		mknod /dev/ttyM0 c 3 128
		mknod /dev/ttyM1 c 3 129

	A better way to do this would be to modify the /etc/conf/node.d/asy
	file so that it creates the modem control ports.  

14. How do I setup a global environment variable that is set for 
all programs/shells (including the /etc/rc*/* scripts?

	There are two ways to do this, depending upon your OS.  
	
	For SCO UNIX (starting with version 3.2v2, I believe), you use the
	file /etc/initscript.  All environment variables and running of
	various environment related utilities (like ulimit, umask, etc)
	should be in this file.  The only exception is the setting of the
	TZ variable, which still should be in /etc/TIMEZONE (for compatibility
	with other existing software).  Note that /etc/initscript does not
	have the same 5 variable limit that is present for the TIMEZONE
	file.

	Note the warning below about changing /etc/TIMEZONE, since it also
	applies to this file.

	For all of the other UNIX systems:

	/etc/init reads the /etc/TIMEZONE file at start up time.  The
	first 5 or so variables in this file are placed into the global
	environment for all processes started by init (this includes the
	rc scripts, cron scripts, etc).  In addition many OS scripts
	will "source" this file.

	NOTE: when modifying the /etc/TIMEZONE file be very careful.  An
	invalid entry could cause all of the gettys initiated by init to 
	fail.  This doesn't happen until the next time you reboot and by
	then you may forget what you changed.

15. How do I get more than 8 virtual terminals to work on the console?

	SCO provides 10-12 MultiScreens(TM, I believe).  Depends on the
	keyboard and how much memory is available.  This is configured
	using the "configure" script.

	Interactive (and probably the other 386 Unix ports) come set up to
	use up to 8 VTs while the driver will support up to 15.  To make
	the changes to allow the use of more than 8 do the following:

	Assuming you want new_max number of vt's (new_max must be <= 15) you can
	do the following....

	a.  Modify the NKDVTTY line in /etc/conf/cf.d/mtune to look like:

		NKDVTTY       	8	8	new_max

	b. Add the following entry to /etc/conf/cf.d/stune:

		NKDVTTY		new_max

	c. reconfigure the kernel

	d. add new devices to /dev.  The device names and numbers should be as 
	   follows:

	   Major device number:	Value in 6th column of the "kd" entry in 
				/etc/conf/cf.d/mdevice file.  (On my system
				it is a 5).

	   Minor device numbers are generated using the following mechanism:

		/dev entries	Minor Device Number
		------------    --------------------------
		vt00 - vt07 	(vt number) * 32) 
		vt08 - vt14 	(vt number - 8) * 32) + 16

		where vt number is the two digits in the device name

	e. Add the appropriate entries to /etc/conf/cf.d/init.base

	f. If you want to see "VT#" on the login prompt for each VT, you must
	   add entries to /etc/gettydefs for each vt.  See the entry for VT02
	   for an example.

	g. reboot the system

16. Why do I get the error "ps: unlink() error" every once in a while?

	This error is caused by the ps program being unable to remove
	the /etc/ps_data file.  This is usually a permissions problem
	on /etc or /bin/ps.  The files and/or directories should
	be set with the following modes:

	drwxrwxr-x  20 root     sys         3968 Apr 29 08:36 /etc
	-r-xr-sr-x   1 root     sys        21472 May 22  1989 /bin/ps

	In english:
		/bin/ps must be mode 2xxx (set-gid)
		/bin/ps must have the same group as /etc (usually sys)
		/etc must be group writable

17. How do I set up more than 4 unix partitions on the disk drive

	Under 386/ix (and Esix) a single "fdisk" partition can contain
	up to 16 unix partitions (including swap, alternates, and reserved).
	However, the default disk configuration software will only set up
	up to 4 partitions for use as file systems. 

	For ISC UNIX (version 2.2 and above), the installation software
	will allow you to specify a full range of partitions.  SCO UNIX
	has a similar capability in its divvy software.

	For those running 386/ix the problem can be bypassed
	using the following manual procedure:

	This is slightly different if you are talking about the 
	boot drive as opposed to the second drive. 

	For the boot drive, use the installation software to set up
	the root (and /usr, if you are combining them) partition as
	you need it to be.  Then place the rest of your disk into
	the last partition.

	For the second drive, use the system administration shell
	to set up the disk and set up the first partition as you 
	want it to end up.  Place the rest of the disk into a single
	partition.

	The rest of this applies to either disk.  I will show you the
	steps required to add 3 partitions on the first drive.

	a. Make a backup of your system!!!

	b. Make sure you made a backup of your system!!!

	c. modify the /etc/inittab so that your system boots up into
	   single user mode.  To do this you change the initdefault 
	   line to be something like:

		is:s:initdefault:

	   Note the 's' between the first set of colons.

	d. reboot the system

	e. Look at the /etc/partitions file and determine the name
	   of the big partition you created.  For my system this was:

		usr2:
		partition = 3, start = 129808, size = 1187914,
		tag = USR, perm = VALID

	   Since we want to break that partition up into smaller
	   partitions, we need to delete it from the disk.
	   the remainder of the disk.  The following command is used:

		/etc/mkpart -p usr2 disk0

	   The "disk0" comes from the first line in the /etc/partitions
	   file which contains the header information for the disk drive.  If
	   you are working on the second disk drive you will find a second
	   device stanza header just like the one for disk0: somewhere in
	   the middle of the file.

	   When you run mkpart you will probably get some message saying
	   that your changes won't take effect until you rebbot.  This is
	   ok.  You don't need to reboot yet.

	f. Now you must edit the /etc/partitions file to set up the
	   partition stanzas for your new partitions and delete the
	   stanza for the /usr2 partition. If your /usr is not part of
	   your root, you might want to mount /usr so that you can 
	   edit the files with vi.  I added the following entries:

		a:
		partition = 3, start = 129808, size = 243600,
		tag = USR, perm = VALID

		b:
		partition = 4, start = 373408, size = 311978,
		tag = USR, perm = VALID

		c:
		partition = 5, start = 685386, size = 469768,
		tag = USR, perm = VALID

		d:
		partition = 6, start = 1155154, size = 162568,
		tag = USR, perm = VALID

	   If some partitions already exist (like alternates or 
	   reserved) you must skip these and add your entries with
	   partition numbers above the existing partitions.  If 
	   you do add entries > partition 7, you must mknod the 
	   extra disk devices as follows:

		mknod /dev/dsk/0s8   b 0 8
		mknod /dev/dsk/0s9   b 0 9
		mknod /dev/dsk/0s10  b 0 10
		mknod /dev/rdsk/0s8  c 0 8
		mknod /dev/rdsk/0s9  c 0 9
		mknod /dev/rdsk/0s10 c 0 10

	   The key parts of these entries are the label, partition
	   number, start, and size.  Be sure that your values do not
	   collide and that the start of each successive partition is
	   equal to the summation of the start of the current partition
	   and its size.

	g. Now to make all of the new partitions run the following
	   commands:

		/etc/mkpart -P a disk0
		/etc/mkpart -P b disk0
		/etc/mkpart -P c disk0
		/etc/mkpart -P d disk0

	h. reboot the computer.  Use the command "uadmin 2 0" to 
	   halt the system from single user mode.

	i. Make the file systems for each of the partitions:
	   
		mkfs /dev/rdsk/0s3 243600
		mkfs /dev/rdsk/0s4 311978
		mkfs /dev/rdsk/0s5 468768
		mkfs /dev/rdsk/0s6 162568:63000
	   
	   Note that since the /d partition is to be my news partition
	   I have overridden the default number of inodes created on 
	   the filesystem.

	j. Label the file systems:

		labelit /dev/rdsk/0s3 a disk0
		labelit /dev/rdsk/0s4 b disk0
		labelit /dev/rdsk/0s5 c disk0
		labelit /dev/rdsk/0s6 d disk0

	k. Modify /etc/fstab.  Delete the entry for /usr2 and add
	   entries for the new file systems.  My new /etc/fstab file
	   looks like:

		/dev/dsk/0s3	/a
		/dev/dsk/0s4	/b
		/dev/dsk/0s5	/c
		/dev/dsk/0s6	/d

	l. Make the mount points:

		mkdir /a /b /c
		
	m. If desired, change the /etc/inittab file initdefault 
	   entry back to what it was before.

	n. run "telinit 2" or "init 2" (or 3 for those of you
	   with a network)

	o. Make a lost and found directory and create some empty slots
	   (for fsck to use when fixing the file system) for each partition.
	   I usually run something like the following:

		for i in a b c d
		do
			mkdir /$i/lost+found
			cd /$i/lost+found
			for j in 1 2 3 4 5 6 7 8 9
			do
				for k in 1 2 3 4 5 6 7 8 9
				do
					> $j$k
				done
			done
			rm /$i/lost+found/[1-9][1-9]
		done

	All done.

18. Why do I run out of inodes on some filesystems when I know I don't
    have that many files?

	It's a known bug that is most often manifested by running
	bnews for news processing (Cnews doesn't seem to cause the
	problem - we have been running Cnews for over a year with a 
	full newsfeed and have never seen the problem).  Binary patches 
	have been posted to this group for several of the SVR3 OSs.  

	It is not know if the problem still exists in the SVR4 products.

	If you have it, write hate mail to your supplier's expensive QA
	department... It has been known for years.

19. What is the minimum hardware for running Unix on a 386?

	A 386SX, 4 MB RAM, 40MB Disk, monochrome adaptor and monitor.
	You can have this for less than $1000 mail order nowadays.
	Using RLL is advisable, and does not cost you more.

	With Xenix, you can live with as little as 2MB.

	You will need additional space for any packages that you wish
	to install (especially the development system and X window 
	packages).

20. Will a caching controller be a good investment?

	The question is not easy to answer. Some believe that
	you'd do better by just adding the same memory to central
	memory and expand the buffer cache. On many machines that are
	already at the limit of 16 MB this cannot be done and so 
	the caching controller becomes a good option.

	WARNING: Caching controllers that are not write-through must
	be protected by a UPS or you could find your system horribly
	trashed when power is lost during heavy disk i/o.

21. How many users can I run on my 386?

	On a suitably configured (read: multiple fast discs, ESDI
	or SCSI controller, many MBytes of RAM), quite a lot,
	depending on many factors. At least a dozen. As many as
	as three dozen doing light sw developement or similar things.

22. What is the bandwidth of an AT Bus?

	Over 5 MBytes per second. As an IO-memory bus, that's still
	plenty. Most peripherals don't go over 1 MByte per second.
	As a CPU-memory bus, no way.

23. Can Unix make use of memory in 16 bit AT bus slots?

	Yes, but the slowdown is really significant. (for two 
	reasons: 1. each 32 bit word access will require two bus
	cycles to obtain since the bus is only 16 bits wide and 2) the
	bus usually only runs at 8MHZ compared to at least 16MHZ for
	the standard memory bus. Don't do it.

	When you buy your machine, make sure that you can put in
	1MBit chips, and that you can stuff at least 8MBytes on
	the motherboard or the 32 bit memory board without buying
	anything extra, and that you can go all the way to 16 MBytes
	(or what you see as your maximum) without throwing away
	any chips that are already installed.

24. How do I fix the error "Out of DOS Inodes" ?

	This error message is telling you that the kernel driver for 
	the DOS file system is out of inodes.  This configuration parameter
	is NDOSINODE, but it is not in the default setup for 386/ix.  To
	fix the problem do the following:

	1. Run /etc/kconfig
	2. Select "configure a kernel"
	3. Select add tunable parameters
	4. Specify NDOSINODE
	5. Specify 500

	It will tell you NDOSINODE is a new parameter and ask you for 
	the minimum, maximum and default.  Acceptable numbers are: 100,
	1000, 200, respectively.

	To avoid kernel compilation warnings about NDOSINODE being
	re-defined you sould edit the /etc/conf/pack.d/dos/space.c file
	and move the line:

	#define	NDOSINODE	200		/* DOS-FSS inodes array */

	to be after the #include of config.h.  You should also add a #ifndef
	NDOSINODE before that line and a #endif after it.

	So that section of the file will now look like:

	#define MINSCTRSHFT     9               /* Minimum sector shift */
	#define MAXSCTRSHFT     10              /* Maximun sector shift */
	#define	NDOSFILSYS	5		/* DOS-FSS superblocks array */

	#include "config.h"	/* for overriding above parameters */
	
	#ifndef NDOSINODE
	#define	NDOSINODE	200		/* DOS-FSS inodes array */
	#endif /* NDOSINODE */

	unsigned char minsctrshft = MINSCTRSHFT;

25. Where can I get the K-Shell (aka ksh)?

	There are several sources where one can obtain a working ksh for 386
	based systems.  These include the following:

		* get source from the AT&T toolchest (approx $3,000 for a full
		  site license)
		* get the MKS trilogy (approx $130) which includes a home-grown
		  substitute ksh.
		* get gnu-BASH which has lots of the functionality and is free.
		* get the KSH from Aspen Technologies

	Both Microport and SCO UNIX (starting with 3.2v2) ship a ksh with
	the standard OS.  The base port of System V R4 includes the ksh as
	one of its standard shell, so if you wait a bit, it will be part
	of the base system.

26. What dos-under-unix product will work with ESIX?

	ESIX does not include a dos-under-unix product.  However, ISC's
	release of VPIX can be installed an run under ESIX.

27. How do I correctly configure the various STREAMS parameters?

	The "STREAMS" subsystem provides a mechanism for inter-process
	communications both within the same system and across various
	networks between different systems.  In the standard OS, pieces like
	X-windows, TCP/IP, NFS, FTP all use STREAMS for communication 
	purposes.  That is why it is important that you correctly configure
	the tunable parameters dealing with STREAMS.

	Those parameters include: data blocks, queues, and streams.  The
	System administration guide will explain what these parameters are,
	but it doesn't explain how to select good values for these parameters.

	To configure these parameters you need to let you system run for
	a while, then run crash(1M) and enter 'strstat' (On ISC UNIX you can
	get the same info much faster with "netstat -m").  For Xenix, similar
	information is available from sw command. The crash(1M) output will
	look similar to the following (Note that I have cut out some unneeded
	columns to get it to fit):

		 ITEM         CONFIG   ALLOC    TOTAL     MAX    FAIL
		streams         512       49      582      75       0
		queues         2048      276     1686     420       0
		message blocks 4440      145    19817     272       0
		dblock totals  3552      145    92323     272       0
		dblock    4     512        0    59276       4       0
		dblock   16    1024       14    24306      37       0
		dblock   64    1024       10    96447      34       0
		dblock  128     512      113    45760     196       0
		dblock  256     256        0    25491       3       0
		dblock  512     128        8    27522      12       0
		dblock 1024      32        0     5303       3       0
		dblock 2048      32        0     8218       7       0
		dblock 4096      32        0        0       0       0


	In looking at this output, the key factors are CONFIG, MAX, and
	FAIL.  The ideal is to keep the CONFIG numbers approx 20% higher
	than your MAX (which will keep FAILs at zero).  Of course, the
	numbers represented under max an fail columns are only since
	the last time you rebooted, so you should keep track of the
	highest values you have seen.

	Once you determine the correct values for your system, you can
	use idtune or kconfig to change the parameters (or edit the
	/etc/conf/cf.d/stune file), rebuild the kernel and see how 
	things improve.

----------------------------------------------------------------------------
If you have suggestions or corrections for any of these answers, please send
them to ...!uunet!virtech!cpcahil.


-- 
Conor P. Cahill            (703)430-9247        Virtual Technologies, Inc.
uunet!virtech!cpcahil                           46030 Manekin Plaza, Suite 160
                                                Sterling, VA 22170 



More information about the Comp.unix.sysv386 mailing list