Query execution time profiling and optimization

[wangli1426]

A query can take over 100 ms even through it does not involve in any subquery on the B+tree or the data chunk.

To reduce the query execution latency, we should first breakdown the query execution time, pinpoint the performance bottlenecks and make optimizations.

[wangli1426]

Via investigation, I found that a large amount of time is spent on filtering on the time stamp and running user-defined filters. Since the data is serialized before inserted into the B+ tree, any filtering on non-key attributes need to deserialized the tuple first, resulting in expensive deserializing cost.

To solve this problem, we need to redesign the implementation of B+ tree. To be more specific, data tuples should not be serialized before flushing the B+ tree. This avoids deserialization in subquery evaluations.