BUGs in Zipfian generator

[Losk-x]

ZETAN is not a constant

ALEX/src/benchmark/zipf.h

Line 10 in 4370da6

static constexpr double ZETAN = 26.46902820178302;

Though this line is directly copied from ycsb, it neglects the fact that ZETAN is determined by the num_keys_ and ZIPFIAN_CONSTANT.
In ycsb's implementation, the num_keys_ is also fixed, so the ZETAN can be a constant. See:

https://github.com/brianfrankcooper/YCSB/blob/dbe92a1fbd24be061fcf01e56838afb83e6f9c58/core/src/main/java/site/ycsb/generator/ScrambledZipfianGenerator.java#L35

https://github.com/brianfrankcooper/YCSB/blob/dbe92a1fbd24be061fcf01e56838afb83e6f9c58/core/src/main/java/site/ycsb/generator/ZipfianGenerator.java#L244

Hash function's keyspace may be too small

ALEX/src/benchmark/zipf.h

Line 60 in 4370da6

inline uint32_t fnv1a(int fourBytes, uint32_t hash = SEED) {

Using a fnv1a-32 is not consistent with yscb (fnv1a-64). It may cause hash collision and make keyspace not fully-covered when num_keys_ is closed with 2^32-1.

[ZEMINGMA]

已收到，感谢来信。祝好！

[Losk-x]

Test result of the wrong generator in 200M dataset:
10B keys are generated, and only 62.5% can be covered.