seriously? I'm always confused by the large number of different low level key-value stores on offer. All claiming to be the fastest, with benchmarks to back it up.
Reading the article, it seems that they could have contributed some development to Rocksdb and saved themselves a lot of time rather than implementing their own system.
RocksDB has the same problems as LevelDB (uses large numbers of file descriptors, considerable read overhead compared to LMDB), plus it's even more complex and harder to configure. I had high hopes for the LSM tree from SQLite4[1], but last I checked development was stalled.
Reading the article, it seems that they could have contributed some development to Rocksdb and saved themselves a lot of time rather than implementing their own system.