[tatersolid]

Don’t forget column-stores. Lists of floats and ints are generally “full size” but then XORed with the preceding value then the result is compressed in chunks with a high-speed compressor like LZ4. We see 20x or higher compression ratios on many of our column-stores which contain data such as invoice amounts as metrics. Decompression overhead is below zero (much faster than operating on the raw uncompressed data which is bound by memory bandwidth or IO)