Hi, I am looking for advice as to what data structure I should use to implement a program with the following characteristics.

My input list will be an unordered list of of 'a, millions of items.
My output structure (the one I am raising the question about) will be some sort of dictionary with a key 'b and the value will be a list of 'a.

e.g. suppose I have [("A",1,7) ; ("B",2,3) ; ("C",1,4) ; ("A",2,3) ; ("C",2,2)] and I want the key to be the first item of each tuple and the value will be the tuple itself, i.e.

1
2
3

"A" -> [ ( "A", 1 , 7 ) ; ( "A", 2 , 3 )]
"B" -> [ ( "B", 2 , 3 ) ]
"C" -> [ ( "C", 1 , 4 ) ; ( "C" , 2 , 2)]

The reason I am bothering to ask about this (which is obviously a simple looking data struct) is because the source list will be processed in parallel and speed is the most important thing for this algorithm so I want to be able to safely run it on a many-cpu server. What I want to know is if there is a standard way, or indeed an existing standard function, that can do something like this...

1

dict.AddOrInsert(key,val)

....specialised for "key" to "list of val" such that the amount of locking is the absolute minimum for the above operation.