That article is nicely written and covers everything. Regarding the specific queries:
Not sure where did you see '-p' ?
1) Do not use 'system power off', this command is available to SP only.
"system power off" from Service Processor (SP) CLI is not a substitute for a graceful shutdown using ontap 'halt'. Therefore, stick to 'halt' command for graceful shutdown.
2) Yes, it is advisable to do 'halt' from the SP CLI of each Node: So that you don't drop the session and you can view the complete sequence of shutdown procedure until it shows the LOADER> prompt, which is the indication that you are SAFE to physically power-off the Controllers. Once Controller is off, you can simply shutdown the the Disk Shelfs. Order of Node doesn't matter for 2-node (for more than 2-node, halt the one holding epsilon last), b'cos you are shutting down the system anyway and I suppose there is no intention to 'keep serving data from other Nodes'. If this is a planned shutdown that means your clients must have been notified already about the unavailability of services during the shutdown period. If the Node is serving cifs, then it makes sense to terminate cifs sessions, prior to running halt.
Firs thing:
1) Note down the SP IPs:
Ontap
::> sp show
2) ssh to SP IP of each Node:
login as: admin
admin@10.x.x.x's password:
BMC cluster-01>
BMC cluster-01> system console
Type Ctrl-D to exit.
SP-login: admin
Password:
*****************************************************
* This is an SP console session. Output from the *
* serial console is also mirrored on this session. *
*****************************************************
cluster::>
