Yes I agree we have to partition function but my concern was with the key we consider for the partition function should be unique/ or part of the primary key..its recommended. My question is why it is should be unique/ column best candidate for Clustered Index if Index is not there ?
The partition function can be anything at all. It can be related to the primary key if that makes sense but can be something else entirely. It's not so much that you choose one "key we consider for the partition function". You choose some column or combination of columns that makes sense for your business. One popular choice is a datetime column if it is set at insertion time to the current datetime (e.g. getdate()) Then your function can say "split on the year 2013", or something.