I thought of those, that's why I was asking if someone had more info on their performance. If I use an array, I have to add in code to check what the last used element was, in order to add new values on to the end. If I use a collection (ordering isn't important for the data) then I have the overhead of a linked list. With the string being essentially an array of char[] I was pretty sure it would come down to either the string or just an array if there was enough overhead removal to warrant the code change. From the link I was given, it looks like that fastest way to go is going to be a 48 element array. Since I'm only dealing with at most 4 of them at a time, they should stay in cache. I only have to keep track of how many elements are there during the adding process to prevent having to scan the array before every addition. For reading I have an easy cheat, since 0 isn't a valid value, I'll know when I hit the last element. What I think will also help out a lot is some restructuring of the class that's being worked with to produce the number lists in the first place. It's a small class to begin with and I think I can reduce it down to a value type struct, by moving the functions into the container class. If all goes well, the whole mess should fold up into cache with a bit of room to spare. Then it should just come down to figuring out how many sets I can queue up from the database at a time.
The true man wants two things: danger and play. For that reason he wants woman, as the most dangerous plaything.
