If you are looking at Quasar compression, you probably have a lot of data.

Not “a large table in Postgres” data. The kind of data that keeps growing, costs real money to move, and still needs to stay online because people query it.

Compression matters. It can turn an impossible workload into a manageable one.

General-purpose compression such as gzip, Zstd, or LZ4 helps, but the gains are rarely transformative at large scale. When they are, the CPU cost may not be compatible with the ingestion or query workload.

Time-series compression algorithms such as delta or delta-delta can perform well, but real-world data defeats them easily. Add noise, gaps, jumps, irregular timestamps, or regime changes, and the compression ratio is often not what you were promised.

This is why we built Delta4C: a compression algorithm optimized for numerical data, able to shift strategies as the data changes.

Delta4C is Quasar’s lossless compression engine for numerical data. It works on integers and floating-point values, and recovers the original data bit for bit.

The core idea is simple: do not assume one compression strategy will fit the data.

Delta4C first analyzes each block quickly enough to stay compatible with live ingestion. It looks at value distributions, deltas, repetition, entropy, and bit-level structure, including floating-point layout.

Then it splits decisions by block, so one noisy region does not force the wrong strategy on the rest of the dataset.

Finally, Delta4C composes the compression pipeline dynamically. A block may benefit from delta or delta-delta encoding, run-length encoding, bit-packing, dictionary compression, floating-point transcoding, or a combination of stages. Delta4C applies the stages that help and skips the ones that do not.

On realistic telemetry, Delta4C delivers roughly 8-12X higher compression than general-purpose compressors. On highly structured data, it can deliver orders-of-magnitude gains while sustaining multi-GiB/s throughput.