While Graph Engine definitely looks useful, I agree that the comparison is unfair. GE is an optimized implementation while associative containers (unordered_*) in C++ are more generalized implementations.
The above blog provided clearly explains optimized implementations of associative containers in C++. It would be interesting to see how these compare with GE.
Please note that this test also only captures a single profile of Graph Engine. The cell entry lookup table and the MM module are designed to operate in a more complex environment and not optimized towards the direction of this micro benchmark(concurrency, logging, a strong type system, to name a few). Of course, it would be very interesting to test against other options, especially those concurrent dictionaries combined with custom allocators.
