shutdown -h now
Shuts down now, halts and returns to the machine’s console prompt.
shutdown -r now
Shuts down now and reboots.
reboot
Same as the above.
You can also use shutdown to specify a time for shutdown,
but this is of limited use.
Flushing the cache
If you ever have to turn off a system running UNIX, it is essential to flush the cache. UNIX uses write-behind caching: you might think you wrote a file to disk, but chances are it’s still in RAM. Therefore, switching off the power makes a mess of UNIX.
Type sync. Wait.
When the prompt returns, assuming no other processes (i.e.
users) write to disk, you can switch off in relative safety.
Broadcasting messages to people
If you want to shut down and there are people on the system, it is courteous to tell them.
wall
Short for “write all”. You need to be root (superuser). Type in your message, press ^D to finish (or ^C to abort).
write user
The same, but to a particular user.
cat > /dev/ttyxx
The rude and amusing way to do it. If you know what terminal
they’re logged on at (see Managing Processes and Snooping below), and you
are superuser, you can write directly to their terminal. That way you don’t
get the beep and the message saying “Message from …”. It can really confuse
people :-)
Which machine am I on?
Type hostname. If it’s not what you expect, don’t shut
it down!
In summary, after testing out of curiosity than for any practical purpose, these are the methods we found, listed in order from slow and safe to fast and dangerous for stopping a UNIX machine.
1) shutdown -h now
2) shutdown -fh now
Some systems use "-F" instead of "-f". Is supposed to skip some safety measures and therefore to be faster.
3) sync ; sync ; halt
Will be even faster but might (only in extreme cases) damage the system because it stops without anything being closed in order. Still, most systems will survive this and boot more or less properly after being killed that way.
4) halt -q
Even faster, but don't try this without a journaling file system (don't try this at all without the need to do so). Old SunOS versions, for instance, had a fair chance of not coming up properly again after such a nosedive. Even with a journaling FS there is some risk, which is why it should only be used as either a very extreme measure or for testing purposes on expendable (test) systems. The same goes for the "winner":
5) cat /etc/hosts > /dev/kmem
Any other file will do too - i use this just out of habit. On POWER and PC hardware (and probably on most other hardware too) this was consistently faster than even "halt -q", it takes only about a dozen bytes to be written over the kernel image to crash the system. In AIX this is not possible any more, because even root cannot write to /dev/kmem in newer versions.
