Tracking issue for improving benchmarking

[LtdJorge]

When making decisions around algorithms and implementations, it is often necessary to prove one choice is better than another by benchmarking them, instead of just making assumptions.

There are benchmarks currently in Limbo, but there is lots of room for improvement. I will list here a general set of requirements for benchmarks that give us a good understanding of how Limbo performs and what can be improved.

• TPC-H support TPC-H support #36
• TPC-E support TPC-E support #685
• Add benches with queries stressing CPU-bound operations (aggregates, joins, etc)
• Add benches with queries stressing IO (full table scans for huge tables, concurrent random IO accesses...)
• Add benches with high query/IO concurrency
• Add benchmarks for the C API that compare SQLite and Limbo