LevelDB benchmarks against Kyoto Cabinet and SQLite
Holy shit – who the hell tests DATABASE by writing thousands-to-millions records of 100 bytes?
There is a test, where they run 1000 values of 100k each – a bit overkill for database, but yet it is still ALL IN MEMORY.
Those tests are actually “how fast we can read from / write to RAM if our control structure is good enough not to content on locks and the like”
What is really interesting is how it behaves, when database size does not fit memory, or at least is close enough, but not hundred of MB.
Yfrog presentation about HBASE usage Sometimes I think I’m doing something wrong
Comments are currently closed.
FYI, I read their benchmarks and came to the same conclusion. If they wanted to reflect reality, they would set cache to 10% of data size, not 100%. One may as well benchmark against std::map.