All articles
Articles.
What we’re learning from compressing real archives and streams, the news from Compress, and case studies with the numbers.
Flowershift launches Compress
Compress, Flowershift’s neural compression product, is now available to research, government and enterprise teams in Australia.
What an error bound actually guarantees
“Lossy” worries scientists for good reason. Here’s what a hard error bound promises, how we check it on every file, and why it beats “looks fine”.
Shrinking a national climate archive
How a climate research centre cut its archive from 2.1 PB to 140 TB with error-bounded compression, and kept every result reproducible.
Learned codecs against zstd, SZ and ZFP
We ran the same ten datasets through general-purpose, scientific and learned compressors. Where neural codecs win, and where they don’t.
GRIB2 and Zarr v3 support in the Compress SDK
The latest SDK reads and writes GRIB2 and Zarr v3 directly, with chunked reads from cloud storage.
Compressing on orbit: more scenes through the same pass
A ground station pass is a fixed window. Compress on board and the window holds more. What it takes to run a codec on a satellite.
More scenes through the same pass
How a satellite operator ran Compress on board to bring down six times more imagery per ground station pass, without new ground stations.
The real cost of keeping a petabyte
Storage is the line item everyone sees. Replication, egress, backups and the time to read it back are the ones that add up.
Research partnership on compressing climate model output
Flowershift is working with a university climate group to test error-bounded compression on a national model archive.
Will your files still decode in ten years?
A learned codec is a model, and models change. How we version every codec so a file written today opens in 2036.
Checkpoints that don’t fill the cluster
How a supercomputing facility shrank simulation checkpoints tenfold, cut write times and kept more runs online for its researchers.
Sensor data over thin links
Cellular, LoRa and satellite IoT all charge by the byte. How compressing at the sensor turns hourly readings into minute-by-minute ones.