> Parquet and Arrow are complementary technologies, and they make some different design tradeoffs. In particular, Parquet is a storage format designed for maximum space efficiency, whereas Arrow is an in-memory format intended for operation by vectorized computational kernels.

> The major distinction is that Arrow provides O(1) random access lookups to any array index, whilst Parquet does not. In particular, Parquet uses dremel record shredding, variable length encoding schemes, and block compression to drastically reduce the data size, but these techniques come at the loss of performant random access lookups.