Currently, we cascade runlength encoding on top of dictionary encoding for string columns. Even if dictionary encoding is not used, the lengths array is encoded using runlength encoding automatically. This is not reasonable for fast storage devices such as SSD.