Temporal counters inaccuracy
- Compression happens during destage, and so data in the write cache is not yet compressed (not reflected by the volume / filesystem "savings" field.
- On a large system with 200TB of data, with a full cache (135GB) this is less than 0.07% of inaccuracy (less for systems with more data)
- Since data in write cache is not compressed, the host write is acknowledged without delay
Compressing pre-existing data
- When upgrading a populated InfiniBox system to a software version that enables compression, the already existing data is not automatically compressed
- Moving old volumes to new compressed volumes (for example, using VMotion) compresses the data
- This limitation will be eliminated in a later release
Writing non-compressed data on a compression-enabled dataset
The InfiniBox system may write data uncompressed for a compression-enabled dataset, if:
- The system wrote the maximum amount of compressed data it can (for data that is highly compressible).
This limit is not a hard number but rather calculated based on the memory consumed by the metadata. We estimate average compression up to 1:3 will not consume all the memory.
- A specific volume / filesystem reached the ceiling compression ratio. Compression on that entity will stop to allow other entities to keep compressing more data.
- Compression is ineffective (for example - data was encrypted by the host and doesn't compress) - the system avoids compression (reduces CPU load)