








        SSeettttiinngg uupp tthhee TThhiirrdd BBeerrkkeelleeyy SSooffttwwaarree TTaappee**

                       _W_i_l_l_i_a_m _N_. _J_o_y
                       _O_z_a_l_p _B_a_b_a_o_g_l_u
                 Computer Science Division
 Department of Electrical Engineering and Computer Science
             University of California, Berkeley
                Berkeley, California  94720


     The  distribution  tape  can  be  used   only   a   DEC
VAX-11/780**  with  RM03  or  RP06  disks and with TE16 tape
drives.  We have the ability to make tapes for systems  with
UNIBUS** disks, but such tapes are inherently rather system-
specific, and will not be discussed here.  The tape consists
of some preliminary bootstrapping programs followed  by  one
dump  of  a  filesystem  (see  _d_u_m_p(1)) and one tape archive
image (see _t_a_r(1));  if  needed,  individual  files  can  be
extracted after the initial construction of the filesystems.

     If you are set up to do it, it is a good  idea  immedi-
ately  to make a copy of the tape to guard against disaster.
The tape is 9-track 1600  BPI  and  contains  some  512-byte
records  followed  by  many  10240-byte  records.  There are
interspersed tapemarks; end-of-tape is signalled by a double
end-of-file.

     The  tape  contains binary images of the system and all
the user level programs, along with source and  manual  sec-
tions  for  them.   There  are  about 4200 UNIX|- files alto-
gether.  The first tape  file  contains  bootstrapping  pro-
grams.   The second tape file is to be put on one filesystem
called the `root filesystem', and contains  essential  bina-
ries and enough other files to allow the system to run.  The
third tape file has all of  the  source  and  documentation.
Altogether  the  files  provided on the tape occupy approxi-
mately 40000 512 byte blocks.|=
-----------
*An early version of this paper appeared under the
title ``Setting up  the  Berkeley  Virtual  Memory
Extensions  to the UNIX Operating System'' and, no
doubt, references to this paper by this name exist
elsewhere  in the documentation.  Portions of this
document are adapted from  ``Setting  Up  Unix/32V
Version  1.0''  by  Thomas  B.  London and John F.
Reiser.
**  DEC, VAX, UNIBUS and MASSBUS are trademarks of
Digital Equipment Corporation.
|- UNIX is a Trademark of Bell Laboratories.
|= UNIX traditionally talks in terms of 512 charac-
ter  blocks,  and for consistency across different









                             -2-


MMaakkiinngg aa ddiisskk ffrroomm ttaappee

     This  description  is an annotated version of the `sys-
gen' manual page in section 8 of the UNIX Programmer's  Man-
ual.  Before you begin to work on the remainder of this man-
ual, be sure you have an up to date  manual,  and  that  you
have  applied  all updates to the manual which were provided
with it, in the correct order.

     Perform the following bootstrap procedure to  obtain  a
disk with a root filesystem on it.

1.   Mount the magtape on drive 0 at load point, making sure
     that the ring is not inserted.

2.   Mount a disk pack on drive 0.

3.   Key in at 50000 and execute the following boot program:
     You  may  enter  in lower-case, the LSI-11 will echo in
     upper-case.  The machine's printouts are shown in bold-
     face,  explanatory  comments are within ( ).  Terminate
     each line you type by carriage return or line-feed.

















-----------
versions of UNIX and to avoid mass confusion, user
programs  in the Virtual Vax version of the system
also talk in terms of 512 blocks, despite the fact
that the file system allocates 1024 byte blocks of
disk space.  All user progras such  as  _l_s(1)  and
_d_u(1) speak in terms of 512 byte blocks; only sys-
tem  maintenance   programs   such   as   _m_k_f_s(1),
_i_c_h_e_c_k(1),  _d_u_m_p(1), and _d_f(1), speak to 1024 byte
blocks.  It is true that i/o is most efficient  in
1024  byte  quantities, but it is most natural for
the user to think of  this  as  ``2  blocks  at  a
time.''   In  any  case, packs remain sectored 512
bytes per sector, and at the lowest driver  levels
the system deals with 512 byte disk records.









                             -3-


     >>>>>>HALT
     >>>>>>UNJAM
     >>>>>>INIT
     >>>>>>D 50000 20009FDE
     >>>>>>D+  D0512001
     >>>>>>D+  3204A101
     >>>>>>D+  C114C08F
     >>>>>>D+  A1D40424
     >>>>>>D+  008FD00C
     >>>>>>D+  C1800000
     >>>>>>D+  8F320800
     >>>>>>D+  10A1FE00
     >>>>>>D+  00C139D0
     >>>>>>D+  00000004
     >>>>>>E 50000/NE:A
     ...                 (machine prints out values, check typing)
     >>>>>>START 50000


     The tape should move and the CPU should halt  at  loca-
     tion  5002A.   If it doesn't, you probably have entered
     the program incorrectly.  Start  over  and  check  your
     typing.

4.   Start the CPU with

          >>>>>>START 0


5.   The console should type

          ==

     If  the disk pack is already formatted, skip to step 6.
     Otherwise, format the pack with:

          (bring in standalone RP06 formatter)
          ==rp6fmt
          ffoorrmmaatt :: FFoorrmmaatt RRPP0066//RRMM0033 DDiisskk

          MMBBAA nnoo.. :: 0         (format spindle on mba uunniitt :: 0         (format unit zero)
          (this procedure should take about 20 minutes)
          (some diagnostic messages may appear here)

          uunniitt :: -1      (exit from formatter)
          ==              (back at tape boot level)


6.   Next, verify the readability of the pack via














                             -4-


          (bring in RP06 verifier)
          ==rpread
          ddrreeaadd :: RReeaadd RRPP0066//RRMM0033 DDiisskk

          ddiisskk uunniitt :: 0            (specify unit zero)
          ssttaarrtt bblloocckk :: 0          (start at block zero)
          nnoo.. bblloocckkss ::             (default is entire pack)

          (this procedure should take about 10 minutes)
          (some diagnostic messages may appear here)
          ## DDaattaa CChheecckk eerrrroorrss :: nnnn (number of soft errors)
          ## OOtthheerr eerrrroorrss :: xxxx      (number of hard errors)
          ddiisskk uunniitt:: -1            (exit from rpread)

          ==                   (back to tape boot)

     If the number of `Other errors' is not zero, considera-
     tion  should  be given to obtaining a clean pack before
     proceeding further.

7.   Create the root file system with the  following  proce-
     dure:

          (bring in a standalone version of the _m_k_f_s (1) program)
          ==mkfs
          ffiillee ssyyss ssiizzee:: 7942      (number of 1024 byte blocks in root)
          ffiillee ssyysstteemm:: hp(0,0)          (root is on drive zero; first filsys there)
          iissiizzee == 55007722             (number of inodes in root filesystem)
          mm//nn == 33 550000              (interleave parameters)
          ==                   (back at tape boot level)

     You  now have a empty UNIX root filesystem.  To restore
     the data which you need to boot the system, type

          (bring in a standalone _r_e_s_t_o_r(1) program)
          ==restor
          TTaappee?? ht(1,1)            (1600 bpi, second tape file)
          DDiisskk?? hp(0,0)            (into root file system)
          LLaasstt cchhaannccee bbeeffoorree ssccrriibbbblliinngg oonn ddiisskk..       (just hit return)
          (30 second pause, then tape should move)
          ...
          ==                   (back at tape boot level)

     Now, you are ready to boot up

BBoooottiinngg UUNNIIXX

     Now boot UNIX:

     (load bootstrap program)
     ==boot
     BBoooott
     :: hp(0,0)vmunix               (bring in _v_m_u_n_i_x off root system)










                             -5-


The bootstrap should then print out the sizes of the differ-
ent parts of the system (text, initialized and uninitialized
data) and then the system should start with a message  which
looks (like):

     6611007722++6611008800++7700112200 ssttaarrtt 00xx44bb44
     VVMM//UUNNIIXX ((BBeerrkkeelleeyy VVeerrssiioonn 22..11)) 11//55//8800
     rreeaall mmeemm  == _x_x_x
     aavvaaiill mmeemm == _y_y_y
     ##

The  _m_e_m  messages give the amount of real (physical) memory
and the memory available to user  programs  in  bytes.   For
example, if your machine has only 512K bytes of memory, then
xxx will be 524228, i.e. exactly 512K.

     UNIX is now running, and the `UNIX Programmer's manual'
applies;  references below of the form X(Y) mean the subsec-
tion named X in section Y of the manual.   The  `#'  is  the
prompt from the Shell, and indicates you are the super-user.
You should first check the integrity of the root file system
by giving the command

     ## chk /dev/rrp0a

which abbreviates

     icheck /dev/rrp0a
     dcheck /dev/rrp0a

The output from _c_h_k should look something like:

     icheck /dev/rrp0a
     /dev/rrp0a:
     files      153 (r=111,d=12,b=8,c=22)
     used 1065 (i=27,ii=0,iii=0,d=1038)
     free 6558
     missing       0
     dcheck /dev/rrp0a
     /dev/rrp0a:
          entries  link cnt
     1    0    0


     The diagnostic from _d_c_h_e_c_k is normal, as inode number 1
is reserved for placement of bad blocks, but currently unim-
plemented.

     The next thing to do is to extract the rest of the data
from the tape.  Comments are enclosed in  (  );  don't  type
these.   The  number in the first command is the size of the
filesystem to be created, in 1024 character blocks, just  as
given to the standalone version of _m_k_f_s above.  (If you have
an RM-03 rather than an  RP-06  use  ``43147''  rather  than









                             -6-


``145673'' in the procedure below.)

     ##/etc/mkfs /dev/rrp0g 145673  (create empty user filesystem)
     iissiizzee == 6655449966                 (this is the number of available inodes)
     mm//nn == 33 550000                   (freelist interleave parameters)
     ##/etc/mount /dev/rp0g /usr    (mount the usr filesystem)
     ##cd /usr                 (make /usr the current directory)
     ##cp /dev/rmt5 /dev/null       (skip first tape file (tp format))
     ##cp /dev/rmt5 /dev/null       (skip second tape file (root))
     ##tar xbf 20 /dev/rmt1         (extract the usr filesystem)
     ##cd /                         (back to root)
     ##/etc/umount /dev/rp0g        (unmount /usr)

All  of  the  data  on the tape has now been extracted.  The
tape will rewind automatically.

     You should now check the consistency of the  /usr  file
system by doing

     ## chk /dev/rrp0g

In order to use the /usr file system, you should now remount
it by saying

     ## /etc/mount /dev/rp0g /usr

again.

MMaakkiinngg aa UUNNIIXX bboooott ffllooppppyy

     The next thing to do is to  make  your  console  floppy
into  a  UNIX boot floppy, by placing some files on it using
_a_r_f_f(1).  If you don't have an extra  console  boot  floppy,
this  is  a good time to make one; see _f_l_c_o_p_y(1), and use it
before playing with _a_r_f_f_.

     Once you have a duplicate copy of the console floppy in
the machine, do the following

     ## cd /usr/src/sys/sys/floppy
     ## arff r *

This  will  put  a  copy  of  each  file  in  the  directory
/usr/src/sys/sys/floppy onto the floppy.  You should now  be
able  to  reboot  using the procedures in _b_o_o_t(8).  Try this
after saying

     ## sync

which allows the system to initiate all i/o before you  halt
the CPU.  It is traditional to say












                             -7-


     ## sync
     ## sync

to give the system a little time, and to then reboot.

TTaakkiinngg tthhee ssyysstteemm uupp aanndd ddoowwnn

     To  bring  the  system up to a multi-user configuration
after a boot all you have to do is hit control-d on the con-
sole.   The  system  will then perform /etc/rc, a multi-user
restart script, and come up on the terminals which are indi-
cated  in  /etc/ttys.   See _i_n_i_t_._v_m(8) and _t_t_y_s(5).  To take
the system down to a single user state you can use

     ## kill 1

when you are up multi-user.  This will  kill  all  processes
and  give  you  a  shell  on the console, as if you had just
booted.

     If you wish to change the lines which  are  active  you
can  edit  the file /etc/ttys, changing the first characters
of lines, and then do

     ## kill -1 1

See _i_n_i_t_._v_m_(_8_) for more information.

AAddddiinngg ddeevviicceess

     The UNIX system running is configured to run  with  the
given  disk  and tape, a console, and 8 DZ11 lines.  This is
probably not the correct configuration.  You  will  have  to
correct the configuration table to reflect the true state of
your machine.

     Before you mess with the source for the  system  it  is
wise to make a backup.  The following will do this:

     ## cd /usr/src
     ## mkdir distsys distsys/h distsys/sys
     ## cd sys/sys
     ## cp * /usr/src/distsys/sys
     ## cd ../h
     ## cp * /usr/src/distsys/h

This  allows  you to find out what you have done to the dis-
tribution system by later running  the  command  _d_i_f_f_d_i_r(1),
comparing these directories.

     NN..BB..::   NNoottee   tthhaatt   tthhee   ssyysstteemm   hheeaaddeerr   ffiilleess  iinn
//uussrr//ssrrcc//ssyyss//hh aarree lliinnkkeedd ttoo tthhee ffiilleess iinn  //uussrr//iinncclluuddee//ssyyss..
SSiinnccee    pprrooggrraammss    wwhhiicchh    ddeeppeenndd    oonn    ccoonnssttaannttss   iinn
//uussrr//iinncclluuddee//ssyyss//ppaarraamm..hh  mmuusstt  ccoorrrreessppoonndd  ttoo  tthhee  rruunnnniinngg









                             -8-


ssyysstteemm,, yyoouu sshhoouulldd bbee ccaarreeffuull ttoo nnoott bbrreeaakk tthheessee lliinnkkss..

     There  are  certain  magic  numbers  and  configuration
parameters embedded in various device drivers that  you  may
have  to  change.   The  device addresses of each device are
defined in each driver.  In case you have  any  non-standard
device  addresses,  just  change  the address and recompile.
Also, if the devices's interrupt vector address(es) are  not
currently  known  to  the  system (this is likely), then the
file /usr/src/sys/sys/univec.c must  be  modified  appropri-
ately:  namely,  the proper interrupt routine addresses must
be placed in the table `UNIvec'.  Use the DZ11 as an example
(as distributed, the DZ11 vectors are assumed to be at loca-
tions c0 and c4 (hexadecimal)).

     You will notice that the system,  as  distributed,  has
conditional  code  in  it.   The current Berkeley system, on
``Ernie Co-vax'' is made by defining IDENT  in  the  ``make-
file'' to be

     IDENT=    -DERNIE -DUCB

This  enables the conditional code both for Berkeley and for
this particular machine.  It is traditional to pick  a  mon-
icker  for your machine, and change IDENT to reflect it, and
to then put in changes  conditionally  whenever  this  makes
sense.  You can be guided by the ERNIE conditional code.

     The  system  comes with 4 drivers which we are using: A
KL/DL driver _k_l_._c_, a Versatec printer/plotter  driver  _v_p_._c_,
and  two  copies  of  a  (old  and simpleminded) UNIBUS disk
driver, named _r_m_._c and _r_p_._c_.  These last two are not in  the
most  aesthetically  pleasing of shapes, but are fully func-
tional.  There is also an RK driver _r_k_._c_,  but  we  are  not
using it, and it may need a little work.

     The  disk  and  tape  drivers  for  the MASSBUS devices
(hp.c, ht.c) are set up to run 1 drive and should be changed
if you have more.

     Now,  make  sure you add any new drivers which you have
to the list of DRIVERS here, and to  the  FILES  and  CFILES
variables so they will be compiled and included in listings.
You can also delete drivers which you don't need from  FILES
and  CFILES, or change the code so that nothing will be com-
piled by using a ``#ifndef''.

     The _m_a_k_e_f_i_l_e has several useful entry points:

clean          Cleans out the directory, removing  ..oo  files
               and the like.

lint           Runs  lint  on  the  system;  the  system was
               almost  lint-free  as  sent  to   you.    See









                             -9-


               _l_i_n_t_e_r_r_s  for  the remaining _l_i_n_t when it was
               distributed.

depend         Creates a new makefile  indicating  dependen-
               cies  on  header  files  by  running a search
               through ..cc  files  looking  for  ``#include''
               lines.   Make  sure you format your code like
               the rest of the  system  so  that  this  will
               work.

print          Produces a nice listing of most everything in
               the system directory in a canonical order.

symbols.sort   Creates a new file for sorting symbols in the
               system namelist.  If you have locally written
               programs which use the  system  namelist  you
               can  put  the symbols which they reference in
               _s_y_m_b_o_l_s_._r_a_w and they will be moved at  system
               generation   to   the  front  of  the  system
               namelist for quicker access.

tags           Creates a _t_a_g_s file for _e_x_, to  make  editing
               of the system much easier.

     Before running _m_a_k_e_, you should check the definition of
the constants in /usr/src/sys/h/param.h The constants  NBUF,
NINODE,  NFILE,  NPROC,  and  NTEXT can be changed, and also
TIMEZONE  and  perhaps HZ if you run on 50 cycles.|-  As dis-
tributed, the system is tuned for a  fairly  large  machine.
(There    are   also   tunable   constants   in   the   file
/usr/src/sys/h/vm.h but ignore them for the time being.)

     To generate a new VM/UNIX do

     ## make clean
     ## make depend
     ## make

and when this works

     ## make lint

to discover any residual bugs.

     The final object file (vmunix) should be moved  to  the
root,  and then booted to try it out.  It is best to name it
/newvmunix so as not to destroy  the  working  system  until
you're  sure  it  does work.  It is also a good idea to keep
the old working version around as  /oldvmunix  (and  perhaps
-----------
|- If you change NINODE,  NFILE,  NPROC  or  NTEXT,
then  the programs _a_n_a_l_y_z_e(1), _p_s(1), _p_s_t_a_t(1) and
_w(1) will have to be recompiled.  A procedure  for
doing this is given below.









                            -10-


even a /oldervmunix) to guard against disaster.

SSppeecciiaall FFiilleess

     Next  you must put in special files for the new devices
in the directory /dev using _m_k_n_o_d(1).  Print the  configura-
tion file /usr/src/sys/sys/conf.c.  This is the major device
switch of each device class (block and character).  There is
one  line  for  each  device configured in your system and a
null line for place holding for those  devices  not  config-
ured.   The  essential  block  special  files were installed
above; for any new  devices,  the  major  device  number  is
selected  by  counting  the  line  number (from zero) of the
device's entry in the block configuration table.   Thus  the
first  entry in the table bdevsw would be major device zero.
This number is also printed in the  table  along  the  right
margin.

     The  minor  device  is the drive number, unit number or
partition as described under each device in section 4.   For
tapes  where the unit is dial selectable, a special file may
be made for each  possible  selection.   You  can  also  add
entries for other disk drives.

     In  reality, device names are arbitrary.  It is usually
convenient to have a  system  for  deriving  names,  but  it
doesn't have to be the one presented above.

     Some  further  notes  on  minor device numbers.  The hp
driver uses the 0100 bit of the minor device number to indi-
cate  whether  or not to interleave a filesystem across more
than one physical device.  See _h_p(4) for more  detail.   The
ht  driver  uses  the  04  bit to indicate whether or not to
rewind the tape when it is closed. The 010 bit indicates the
density of the tape on TE16 drives.  Again, see _h_t(4).

     The  naming  of  character  devices is similar to block
devices.  Here the names are even more arbitrary except that
devices  meant  to  be  used  for teletype access should (to
avoid confusion, no other reason) be named /dev/ttyX,  where
X  is some string (as in `0' or `d0').  While it is possible
to use truly arbitrary  strings  here,  the  accounting  and
noticeably  the _p_s(1) command make good use of the fact that
tty names (at Berkeley) are distinct in the first 2  charac-
ters.   In  fact, we use the following convention: ``ttyN'',
with N a number for normal DZ ports. ``ttydX'' with X a sin-
gle  character (starting from 0) for dialups, ``ttykX'' with
X a single letter for  KL  ports,  and  ``console''  (abbrev
``co'') for the console.  This works out well.

     The  files console, tty0-tty7, mem, kmem, kUmem, floppy
and null are already correctly configured,  as  are  special
files  for the default paging are /dev/drum, and the raw and
block versions of the root and /usr file systems.









                            -11-


     The disk and magtape drivers provide a `raw'  interface
to the device which provides direct transmission between the
user's core and the device and  allows  reading  or  writing
large records.  The raw device counts as a character device,
and conventionally has the name of the  corresponding  stan-
dard  block  special  file with `r' prepended.  Thus the raw
magtape files are called /dev/rmtX.

     Whenever special files  are  created,  care  should  be
taken  to  change the access modes _(_c_h_m_o_d(1)) on these files
to appropriate values.

BBaassiiccss ooff DDiisskk LLaayyoouutt

     If there are to be more filesystems mounted  than  just
the  root and /usr, use _m_k_f_s(1) to create any new filesystem
and put its mounting in the file /etc/rc  (see  _i_n_i_t(8)  and
_m_o_u_n_t(1)).   (You  might  look at /etc/rc anyway to see what
has been provided for you.)

     There are several considerations  in  deciding  how  to
adjust  the  arrangement  of  things on your disks: the most
important is making sure there is adequate space for what is
required; secondarily, throughput should be maximized.  Pag-
ing space is an important parameter.  The system as distrib-
uted  has  33440  (512  byte) blocks in which to page.  This
should be large enough for most sites.  You can change  this
if local wisdom indicates this is not good.

     Many  common system programs (C, the editor, the assem-
bler etc.)  create intermediate files in the /tmp directory,
so  the  filesystem where this is stored also should be made
large enough to accommodate most high-water marks.  The root
filesystem  as  distributed is quite large, and there should
be no problem.  All the programs that create files  in  /tmp
take  care to delete them, but most are not immune to events
like being hung up upon, and can leave dregs.  The directory
should be examined every so often and the old files deleted.

     Exhaustion of user-file space is certain to  occur  now
and  then; the only mechanisms for controlling this phenome-
non are occasional use of _d_u(1), _d_f(1), _q_u_o_t(1), threatening
messages of the day, and personal letters.

     The  efficiency  with which UNIX is able to use the CPU
is largely  dictated  by  the  configuration  of  disk  con-
trollers.   For  general time-sharing applications, the best
strategy is to try to split the root filesystem  and  system
binaries  (/usr),  the  temporary  files and paging activity
(/tmp and /dev/drum), and the user files  among  three  disk
arms.  We will discuss such considerations more below.

     Once  you  have  decided  how  to make best use of your
hardware, the question is how to initialize it.  If you have









                            -12-


the  equipment, the best way to move a filesystem is to dump
it _(_d_u_m_p(1)) to magtape,  use  _m_k_f_s(1)  to  create  the  new
filesystem,  and  restore (_r_e_s_t_o_r(1)) the tape.  If for some
reason you don't want to use magtape, dump accepts an  argu-
ment  telling  where  to put the dump; you might use another
disk.  Sometimes a filesystem has to be increased in logical
size  without  copying.  The super-block of the device has a
word giving the highest address which can be allocated.  For
relatively  small  increases, this word can be patched using
the debugger (_a_d_b(1)) and the free list reconstructed  using
_i_c_h_e_c_k(1).  The size should not be increased very greatly by
this technique,  however,  since  although  the  allocatable
space  will  increase  the  maximum number of files will not
(that is, the i-list  size  can't  be  changed).   Read  and
understand the description given in _f_i_l_e_s_y(5) before playing
around in this way.

     If you have to merge a filesystem into another,  exist-
ing  one, the best bet is to use _t_a_r(1).  If you must shrink
a filesystem, the best bet  is  to  dump  the  original  and
restor  it  onto the new filesystem.  However, this will not
work if the i-list on the smaller filesystem is smaller than
the  maximum  allocated inode on the larger.  If this is the
case, reconstruct the filesystem  from  scratch  on  another
filesystem  (perhaps using _t_a_r(1)) and then dump it.  If you
are playing with the root filesystem and only have one drive
the  procedure is more complicated.  What you do is the fol-
lowing:

1.   GET A SECOND PACK!!!!

2.   Dump the root filesystem to tape using _d_u_m_p(1).

3.   Bring the system down and mount the new pack.

4.   Load the standalone versions of _m_k_f_s(1)  and  _r_e_s_t_o_r(1)
     from the floppy with a procedure like:


























                            -13-


          >>>>>>UNJAM
          >>>>>>INIT
          >>>>>>LOAD MKFS
               LOAD DONE, xxxx BYTES LOADED
          >>>>>>ST 2

          ...

          >>>>>>H
               HALTED AT yyyy
          >>>>>>U
          >>>>>>I
          >>>>>>LOAD RESTOR
               LOAD DONE, zzzz BYTES LOADED

          ... etc


5.   Boot  normally using the newly created disk filesystem.

     Note that if you change the disk  partition  tables  or
add new disk drivers they should also be added to the stand-
alone system in /usr/src/sys/stand.

SSyysstteemm IIddeennttiiffiiccaattiioonn

     You should edit the files:

     /usr/include/ident.h
     /usr/include/whoami.h
     /usr/include/whoami

to correspond to your system, and then recompile and install
_g_e_t_t_y_._v_m(8) via:

     ## cd /usr/src/cmd
     ## DESTDIR=/
     ## export DESTDIR
     ## Admin/mk getty.vm.c

This will arrange for an appropriate banner to be printed on
terminals before users log in.

AAddddiinngg NNeeww UUsseerrss

     See _a_d_d_u_s_e_r(8); local needs will undoubtedly dictate  a
somewhat different procedure.

MMuullttiippllee UUsseerrss

     If  UNIX  is  to  support simultaneous access from more
than just the console terminal, the file /etc/ttys (_t_t_y_s(5))
has  to be edited.  To add a new terminal be sure the device
is configured and the special  file  exists,  then  set  the









                            -14-


first  character  of  the appropriate line of /etc/ttys to 1
(or add  a  new  line).   You  should  also  edit  the  file
/etc/ttytype placing the type of the new terminal there (see
_t_t_y_t_y_p_e(5)).  The file /etc/ttywhere is also a useful one to
keep up to date.

     Note  that  init.c  will have to be recompiled if there
are to be more than 100 terminals.  Also note  that  if  the
special  file  is  inaccessible  when init tries to create a
process for it, the system will thrash trying  and  retrying
to open it.

FFiillee SSyysstteemm HHeeaalltthh

     Periodically  (say  every day or so) and always after a
crash, you should check all the filesystems for  consistency
(_i_c_h_e_c_k_,  _d_c_h_e_c_k(1)).   It  is  quite  important  to execute
_s_y_n_c(8) before rebooting or taking the machine  down.   This
is done automatically every 30 seconds by the _u_p_d_a_t_e(8) pro-
gram when a multiple-user system is running, but you  should
do it anyway to make sure.

     Dumping  of  the  filesystem  should be done regularly,
since once the system is going it is  very  easy  to  become
complacent.   Complete and incremental dumps are easily done
with _d_u_m_p(1).  See the scripts /etc/dumpusr  and  /etc/dump-
root  which  we  use  to  dump our file systems.  You should
arrange to do a towers-of-hanoi dump sequence; we tune  ours
so  that  each  file is dumped twice and kept for at least a
week in most every case.  We take fully dumps every  two  to
three  weeks  (and  keep  these  indefinitely).   Note  that
/etc/dumpusr maintain the file /etc/lastdumpdone.  This  can
be  printed out at login by people who can easily then start
dumps if one is needed.

     Dumping of files by name is best done by _t_a_r(1) but the
number  of  files is somewhat limited.  Finally if there are
enough drives entire disks can be copied  with  _d_d(1)  using
the raw special files and an appropriate block size.

CCoonnvveerrttiinngg 3322//VV FFiilleessyysstteemmss

     The  best  way  to convert filesystems from 32/V to the
new format is to use  _t_a_r(1).   After  converting,  you  can
still restore files from your old-format dump tapes (yes the
dump format  is  different,  sorry  about  that),  by  using
``512restor''  instead  of ``restor''.  If you wish, you can
move whole file systems from 32/V to the new system by using
``dump'' and then ``512restor''.

RReeggeenneerraattiinngg tthhee ssyysstteemm

     It  is quite easy to regenerate the system, and it is a
good idea to try this once right away to  build  confidence.









                            -15-


The  system consists of three major parts: the kernel itself
(/usr/src/sys/sys), the user programs (/usr/src/cmd and sub-
directories),  and the libraries (/usr/src/lib*).  The major
part of this is /usr/src/cmd.

     We have  already  seen  how  to  recompile  the  system
itself.   The  three  major  libraries  are the C library in
/usr/src/libc and the FORTRAN libraries /usr/src/libI77  and
/usr/src/libF77.   In  each  case  the  library is remade by
changing into the corresponding directory and doing

     ## make

and then installed by

     ## make install

Similar to the system,

     ## make clean

cleans up.  The source for all other libraries  is  kept  in
subdirectories  of /usr/src/lib; each has a makefile and can
be recompiled by the above recipe.

     Recompiling all user programs is accomplished by  using
a  directory  in  /usr/src/cmd, called Admin, which contains
two files: mk and destinations.  The  file  mk  is  a  shell
script for recompiling files in /usr/src/cmd.  For instance,
to recompile ``date.c'', all one has to do is

     ## cd /usr/src/cmd
     ## Admin/mk date.c

this will place a stripped version of the binary of ``date''
in /usr/dist3/bin/date, since date normally resides in /bin,
and Admin is building a  file-system  like  tree  rooted  at
/usr/dist3.   You  will have to make the directory dist3 for
this to work.  It is possible to use any directory  for  the
destination,   it   isn't   necessary  to  use  the  default
/usr/dist3.  You can set the default by doing:

     ## DESTDIR=_p_a_t_h_n_a_m_e
     ## export DESTDIR


     To regenerate all the system source you can do
















                            -16-


     ## DESTDIR=/usr/newsys
     ## export DESTDIR
     ## cd /usr
     ## rm -r newsys
     ## mkdir newsys
     ## cd /usr/src/cmd
     ## Admin/mk * > Admin/errs 2>& 1 &

This will take about 4  hours  on  a  reasonably  configured
machine.   When  it finished you can move the hierarchy into
the normal places using _m_v(1) and _c_p(1), and then execute

     ## DESTDIR=/
     ## export DESTDIR
     ## cd /usr/src/cmd/Admin
     ## mk ALIASES
     ## mk MODES

to link files together as necessary and to set all the right
set-user-id bits.

MMaakkiinngg oorrddeerrllyy cchhaannggeess

     In  order  to keep track of changes to system source we
migrate changed versions  of  commands  in  /usr/src/cmd  in
through  the  directory /usr/src/new and out of /usr/src/cmd
into /usr/src/old for a time before removing them.   Locally
written  commands  which  aren't  distributed  are  kept  in
/usr/src/local and their binaries are  kept  in  /usr/local.
This  allows /usr/bin /usr/ucb and /bin to correspond to the
distribution tape (and to the manuals that people can  buy).
People  wishing  to  use  /usr/local commands are made aware
that they aren't in the  base  manual.   As  manual  updates
incorporate these commands they are moved to /usr/ucb.

     A directory /usr/junk to throw garbage into, as well as
binary directories /usr/old and /usr/new  are  very  useful.
The   man   command  supports  manual  directories  such  as
/usr/man/manj for junk and /usr/man/manl for local  to  make
this or something similar practical.

IInntteerrpprreettiinngg ssyysstteemm aaccttiivviittyy

     The _v_m_s_t_a_t program provided with the system is designed
to be an aid to monitoring  systemwide  activity.   Together
with  the _p_s(1) command (as in ``ps av''), it can be used to
investigate systemwide virtual activity.  You should  modify
_v_m_s_t_a_t(1m)  so  that  it  prints out disk statistics for the
disks you have, changing the headers to something  appropri-
ate  for your system.  Examine the definitions of DK_UNIT in
the disk drivers supplied to see how the code in _v_m_s_t_a_t  and
_i_o_s_t_a_t and the system correspond.











                            -17-


     By  running  _v_m_s_t_a_t  when  the system is active you can
judge the system activity in several dimensions: job distri-
bution,  virtual  memory load, paging and swapping activity,
disk and cpu utilization.   Ideally,  most  jobs  should  be
either running (RQ) or sleeping (SL), there should be little
paging or swapping activity, there should be available band-
width  on  the  disk  devices  (most single arms peak out at
about 30-35 tps in practice), and the user  cpu  utilization
(US) should be high (about 60%).

     If  the  system is busy, then the number of active jobs
may be large, and several of these jobs may often be in disk
wait  (DW).   If the virtual memory is very active, then the
paging demon may be running (SR will be  non-zero).   It  is
healthy  for the paging demon to free pages when the virtual
memory gets active; it is triggered by the  amount  of  free
memory  dropping below a threshold and increases its pace as
free memory goes to zero.

     If you run _v_m_s_t_a_t when the system is busy  (a  ``vmstat
5'' is best, since that is how often most of the numbers are
recomputed by the system), you can find imbalances by noting
abnormal  job  distributions.  If a large number of jobs are
in disk wait (DW) or page wait (PW), then the disk subsystem
is  overloaded or imbalanced.  If you have a large number of
non-dma devices or open teletype  lines  which  are  ``ring-
ing'',  or  user  programs  which  are doing high-speed non-
buffered input/output, then the system time may go very high
(60-70% or higher).

     If  the  system  is very heavily loaded, or if you have
very little memory relative to your load (512K is little  in
most any case), then the system may be forced to swap.  This
is likely to be accompanied by  a  noticeable  reduction  in
system  performance  as  the  system does not swap ``working
sets'', but rather forces jobs to reinitialize  their  resi-
dent  sets  by demand paging.  If you expect to be in a mem-
ory-poor environment for an extended period you  might  con-
sider administratively limiting system load.

TTuunnaabbllee ccoonnssttaannttss

     There  is  a  modicum of tuning available in the paging
replacement algorithm if it appears to be  badly  tuned  for
your  configuration.  The page replacement (clock) algorithm
is run whenever there are not LOTSFREE pages available (this
and  all  other  constants discussed here are defined in the
system  header  file  /usr/src/sys/h/vm.h).   This  sets  up
resistance  to consumption of the remaining free memory at a
minimal rate SLOWSCAN, which gives  the  desired  number  of
seconds  between  successive examinations of each page.  The
rate at which the clock algorithm is run increases  linearly
to  a desired rate of FASTSCAN when there is no free memory.
Thus as the  available  free  memory  decreases,  the  clock









                            -18-


algorithm  works harder to hold on to what is left.  If less
than DESFREE pages are available  and  the  paging  rate  is
high,  then the system will begin to swap processes out.  If
less than MINFREE pages are available then the  system  will
begin to swap, regardless of the paging rate.

     When  it  has to swap, the system first tries to find a
process which has been blocked for a long time and  swap  it
out  first.   If  there  are no jobs of this flavor, then it
will choose among the remaining jobs in  a  strictly  round-
robin fashion, based on core residency time.  It attempts to
guarantee (during periods of very heavy  load)  enough  core
residency  to  a process to allow it to at least rebuild its
set of active pages (since it must do so by demand  paging).
Processes which are swapped out with large numbers of active
pages similarly receive lower priority for swapin,  favoring
small jobs to return to the core resident set quickly.

     It  is _v_e_r_y desirable that the system run under reason-
ably heavy load with little swapping, with the memory parti-
tioning  being  done  by  the  clock  replacement algorithm,
rather than by the swapping algorithm.  The costs associated
with paging activity are the time spent in the paging demon,
the overhead associated with reclaim page faults  (RE),  and
the extra disk activity associated with pagins and pageouts.
We will discuss disk  considerations  later;  when  kept  to
about  40 reclaim faults per second, the cost of reclaims is
less than 1% of total processor time.  The cpu  time  (shown
by ``ps l2'') accumulated by the pageout demon will show how
much overhead it is generating.

     The system, as distributed, runs the replacement  algo-
rithm  whenever  less  than  1/8 of the total user memory is
free.  This is done starting with  a  30  second  revolution
time  of  the  clock algorithm and increasing to a 20 second
revolution time when there is no free memory.  The goal here
is  to  use  as  much memory as possible (i.e. have the free
list short) but to not have the system run out and start  to
swap.   You can experiment with changing the writable copies
of these variables (e.g. ``lotsfree'' is the  writable  copy
of LOTSFREE) using _a_d_b_, as in:

     ## adb -w /vmunix /dev/kmem
     /m 0 #ffffffff
     lotsfree/D
     ---adb prints value of lotsfree---
     /W 0t100

Here  the ``/W 0t100'' command changed the value of _l_o_t_s_f_r_e_e
to be 100 (decimal).













                            -19-


BBaallaanncciinngg ddiisskk llooaadd

     It is critical for good  performance  to  balance  disk
load.   There  are at least five components of the disk load
which you can divide between the available disks:

     1. The root file system.
     2. The /tmp file system.
     3. The /usr file system.
     4. The user files.
     5. The paging activity.

In our system we currently have 1.75M bytes of memory and  2
disks:  an  RP06  and  an AMPEX 9300.  We run with the root,
/tmp, and paging activity on the RP-06, while the /usr  file
system  and user files are on the 9300.  This gives a fairly
even split of file activity in our environment.

     A split such as this is a good  initial  guess  if  you
have two arms.  If you are fortunate to have three arms, you
can try splitting the  files  up  various  ways.   The  most
important  things  to remember are to even out the disk load
as much as possible, and to do this by decoupling file  sys-
tems  (on separate arms) between which heavy copying occurs.
Note that a long term average balanced load  is  not  impor-
tant...  it  is  much more important to have instantaneously
balanced load when the system is busy.

     Intelligent experimentation  with  a  few  file  system
arrangements  can  pay off in much improved performance.  It
is particularly easy to move the root, the /tmp file  system
and  the  paging  area.   Place  the user files and the /usr
directory as space needs dictate  and  experiment  with  the
other, more easily moved file systems.

     Finally,  when  you have your configuration worked out,
you should set the constant MAXPGIO  based  on  the  maximum
number  of transfers you can expect from your paging device.
The system assumes that if more transfers than  this  occur,
then the system is overloaded, and unless there is a reason-
able amount of free memory, it then  begins  to  swap.   The
swap  scheduler  also consider jobs small if their size when
they were swapped out was less  than  twice  this  constant.
Such small jobs have a better chance of getting swapped back
in when the core situation  is  tight,  since  they  can  be
expected to be able to run in a small number of page frames.

PPrroocceessss ssiizzee lliimmiittaattiioonnss

     As distributed, the system provides for  a  maximum  of
64M  bytes of resident user virtual address space.  The size
of the text, and data segments of a single process are  cur-
rently  limited to 4M bytes each, and the stack segment size
is limited  to  512K  bytes.   These  can  be  increased  by









                            -20-


changing  the  constants MAXTSIZ, MAXDSIZ and MAXSSIZ in the
file /usr/src/sys/h/vm.h.  You must be careful in doing this
that  you  have adequate paging space.  As configured above,
the system has only 16M bytes of paging area.|-

     To increase the amount of resident virtual space possi-
ble,    you   can   alter   the   constant   USRPTSIZE   (in
/usr/src/sys/h/vm.h) and by correspondingly change the defi-
nitions of _U_s_r_p_t_m_a_p and _S_y_s_s_i_z_e in /usr/src/sys/sys/locore.s

     The 4M byte limitation on individual  segment  size  is
enforced  by  the constants MAXTSIZ, MAXDSIZ and MAXSSIZ for
the text, data and stack segments respectively.   These  can
be  increased, given the availability of adequate amounts of
swap space, up to 16M bytes.  In order to increase the  size
of  the  stack or data segments beyond 16M, you will have to
increase the amount of space which can be mapped by the cor-
responding disk map.  To increase, for instance, the maximum
segment size to 32M bytes it would  be  adequate  to  double
both  the  initial  and maximum sizes for the diskmap.  Thus
defining (in /usr/src/sys/h/dmap.h)

     #define   DMMIN          32
     #define   DMMAX     8192

(i.e. a minimum segment size of 16K bytes and a maximum size
of  4M  bytes)  would  allow the 16 diskmap slots to map 32M
bytes.

OOtthheerr lliimmiittaattiioonnss

     Due to the fact that the file system block numbers  are
stored in page table ppgg__bbllkknnoo entries, the maximum size of a
file system is limited to 2^20 1024 byte  blocks.   Thus  no
file system can be larger than 1024M bytes.

     The number of system buffers is limited to be less than
64 because of the way that the MASSBUS adaptor map registers
are  initialized.  The construct ``(128<<9)'' appears in the
/usr/src/sys/sys/mba.c   and   /usr/src/sys/sys/hp.c   code,
silently enforcing this restriction.

SSccaalliinngg ddoowwnn

     If you have less than 1M byte of memory you may wish to
scale the paging system down, by reducing some  fixed  table
sizes  not  directly  related  to  the  paging  system.  For
instance, we have raised NBUF from 32 to 48, NCLIST from 150
to  500,  (also increasing the basic clist block size CBSIZE
from 12 to 28) and NPROC, NINODE  and  NFILE  for  a  fairly
large   system  from  the  way  they  were  distributed  for
-----------
|- Recovery from running out of paging area is cur-
rently  not handled gracefully: the system panics.









                            -21-


UNIX/32V.  We also increased TTLOWAT (from 50  to  125)  and
TTHIWAT  (from  150  to  650.)  If you pull NCLIST down, you
should adjust these also.  You can use _p_s_t_a_t(1m) to find out
how much of these structures are typically in use.  Although
the document is pretty much obsolete for the  VAX,  you  can
see  the  last few pages of ``Regenerating System Software''
in Volume 2B of the programmers manual for hints on  setting
some of these constants.

OOddddss aanndd EEnnddss

     The programs dump, icheck, quot, dcheck, ncheck, and df
(source in /usr/source/cmd) should  be  changed  to  reflect
your  default  mounted  filesystem devices.  Print the first
few lines of these programs and the changes will be obvious.
You  will  probably  want to amend some of /usr/lib/crontab,
and will certainly want to add to /usr/lib/Mail.rc.

     You should periodically examine the file  /usr/adm/mes-
sages, which acts as a system error log (see dmesg(1m)).  In
particular, memory controller errors are checked  for  every
10  minutes  and  a  diagnostic is produced (printing memory
controller register C) if there were any errors.



                                        Good Luck

                                        William N. Joy
                                        Ozalp Babaoglu






























