Why is Map slower than Dictionary on reads, in F# (from the benchmark I have attached)

Question

I stumbled across this page: http://jackfoxy.com/simple-lookups-in-fsharp/

It benchmarks various collections for insertion and retrieval.

If we look at this table (the second on the web page):

String Key by Number of Elements in Lookup Structure, 10,000 random lookups
        Map     IntMap  Dict    HashTbl     HshMltMap
10^2    1.3     n/a     0.4     0.3     1.5
10^3    1.7     n/a     0.4     0.4     1.5
10^4    3.0     n/a     0.7     0.7     1.8
10^5    5.3     n/a     1.5     1.2     2.4
10^6    8.4     n/a     1.6     1.5     6.3

We can see that the lookup with a Map is 5x slower than the Dictionary when it becomes large.

Since Map is read-only and has therefore all the luxury to organize the data in the most optimal way as it doesn't deal with insertions, resizing, etc.. why is it so much slower?

Answer 1

C# Dictionary is implemented using hash tables thus speed of lookup is closed to O(1) .

F# Map, due to the immutability, must use binary trees thus speed of lookup is O(logN) .

F# Map, and other immutable data structures, actually deal with insertions. You still need to insert an element to a map, but get a new map, and the old map is still intact. To make this effective, the new one and the old one must share elements (no cloning/duplicating involved).

And yes, the best way to do that is using trees (F# Map, Set) or linked lists (F# List).