When choosing a file server to run Non-Dedicated NetWare,  it
is   extremely  important  to  choose  a  machine  known  to   run
Non-Dedicated  NetWare.   The following is an summary of why  some
machines will have problems with Non-Dedicated NetWare.

     First,  the  acronym  "ND286" stands  for  Non-Dedicated  286
NetWare in any release from v2.0a through v2.15.   ND286 runs  DOS
applications in 'real mode'  and file server processes execute  in
protected  mode.  Doing  both real and  protected  modes  requires
switching the 80x86 back and forth between the two. Switching from
real  to  protected is accomplished by executing a  special  80286
machine instruction.
     However,  the 80286 does not have a way of directly switching
from  protected to real (the 80386 does,  but ND286 is based  upon
the 80286 architecture).   Usually due to this missing link in the
80286's instruction set, the 80286 is often called a brain-damaged
chip.   Anyhow, the only way to do the switch (from protected mode
to  real mode) is to reset the 80x86.   ND286 relies  on  specific
characteristics  of  the  IBM  AT  hardware  (i.e.  special  reset
hardware)  and  IBM ROM BIOS to perform the  reset  and  correctly
control the switch back to real mode.
     Resetting  the hardware is messy,  and involves some  special
keyboard  processor  and ROM BIOS functionalities.   The  keyboard
processor  is given a command to reset the 80286.   The  ROM  BIOS
needs to be able to recognize at reset time that it isn't doing  a
power on reset (Power On Self Test,  or POST) and instead transfer
control to the  correct real mode address.   The way that the  ROM
BIOS  recognizes that it is doing a protected to real mode  switch
is  via a special 'shutdown byte'.  This byte  contains  different
values,  depending  upon  which  protected  to  real  mode  switch
occurred. 
     The ROM BIOS does not provide any support while the processor
is  running  software  in protected mode,  so  switching  back  is
entirely up to the operating system software.   Any implementation
differences  from  the IBM AT can potentially cause ND286  not  to
work.   (In  doing the reset, NetWare uses shutdown  request  type
0Ah,  which  should  cause  the ROM BIOS to do  a  JMP  DWORD  PTR
40h:67h.)
     The  ROM  should not reset any hardware  (in  particular  the
interrupt  controller) before doing the JMP.   The ROM may  enable
interrupts during processing;  however,  enabling interrupts  will
not make any difference  because NetWare will have masked off  all
interrupts  at the interrupt control chip prior to requesting  the
reset.   NetWare  restores  the interrupt mask  when  it  receives
control back in real mode.
     The  keyboard  processor,   which  is  contained  within  the
keyboard  unit,  is used primarily to give keyboard scan codes  to
the  resident  processor.    Other  functions  that  the  keyboard
processor are used for,  on the AT,  are after thoughts.   One  of
those after thoughts is the request to enable/disable address line 
20 or reset the 80286. 
     The  keyboard processor in at least one clone  cannot  handle
the condition when the keyboard controller wants to give the 80286
another keyboard character.   Instead of the 80286 requesting  the
keyboard character,  it requests an enable/disable address 20 or a
reset  the 80286 command.   It seems that the  keyboard  processor
microcode can only handle giving the next character once it has  a
character,  not these auxiliary requests.   This problem manifests
itself  if the nondedicated file server hangs when a keyboard  key
becomes stuck.
     The  AT keyboard processor is used to manage the  address  20
enable/disable line.  The address 20 logic controls whether or not
the address 20 bit coming from the 80286 is actually asserted onto
the bus.  Disabling address 20 makes the 80286 wrap  segments that
exist beyond 1Mb back into low memory (ala 808x emulation).   This
wrapping  needs  to be enabled for the real to  protected  switch. 
The keyboard processor is used to enable/disable line 20 and takes
1ms  on the IBM AT.    Whereas,  on some clones it can take up  to
50ms (or more!).
     So,  there  are two areas that prevent ND286  to  operate  on
clone machines.   First,  and easiest to correct,  is the  problem
with a sluggish keyboard processor.  To correct this,  replace the
existing keyboard with a true blue IBM keyboard.   There are  some
clone keyboards that also work in place of the IBM keyboard,  but,
unlike  the  IBM keyboard,  not everyone of them  works  in  every
situation.   The second problem is when the BIOS in the machine is
not entirely emulating an IBM AT BIOS.   Usually the culprit  here
is that a Phoenix BIOS is being used.   Of late,  Phoenix seems to
have the problem with their BIOS and ND286 resolved,  but only  in
the  latest revisions of their BIOS.   A proven performer  is  the
BIOS available from AWARD.   When a machine fails to run ND286 and
the keyboard change doesn't cure it,  replacing the machine's BIOS
with  an AWARD BIOS usually reconciles the problem.   However,  be
aware that there are those machines that are such poor clones that
they will never be able to utilize ND286.
     Alas  there is one other problem which may stem from a  clone
that  is  not  a  complete clone.   The  80286  must  be  isolated
correctly from the main memory bus or a DMA being done to or  from
a  network  board  may  become corrupted  during  an  80286  reset
sequence.