Storage Foundation Cluster File System High Availability 7.2 Administrator's Guide - Solaris

About deduplication chunk size

The deduplication chunk size, which is also referred to as deduplication granularity, is the unit at which fingerprints are computed. A valid chunk size is between 4k and 128k and power of two. Once set, the only way to change the chunk size is to remove and re-enable deduplication on the file system.

You should carefully select the chunk size, as the size has significant impact on deduplication as well as resource requirements. The size directly affects the number of fingerprint records in the deduplication database as well as temporary space required for sorting these records. A smaller chunk size results in a large number of fingerprints and hence requires a significant amount of space for the deduplication database.

While the amount of storage that you save after deduplication depends heavily on the dataset and distribution of duplicates within the dataset, the chunk size can also affect the savings significantly. You must understand your dataset to get the best results after deduplication. A general rule of thumb is that a smaller chunk size saves more storage. A smaller chunk size results in more granular fingerprints and in general results in identifying more duplicates. However, smaller chunks have additional costs in terms of database size, deduplication time, and, more importantly, fragmentation. The deduplication database size can be significantly large for small chunk sizes. Higher fragmentation normally results in more file system metadata and hence can require more storage. The space consumed by the deduplication database and the increased file system metadata can reduce the savings achieved via deduplication. Additionally, fragmentation can also have a negative effect on performance. The Veritas File System (VxFS) deduplication algorithms try to reduce fragmentation by coalescing multiple contiguous duplicate chunks.

Larger chunk sizes normally result in a smaller deduplication database size, faster deduplication, and less fragmentation. These benefits sometimes come at the cost of less storage savings. If you have a large number duplicate files that are small in size, you still can choose a chunk size that is larger than the file size. A larger chunk size does not affect the deduplication of files that are smaller than the chunk size. In such cases, the fingerprint is calculated on the whole file, and the files are still deduplicated.

Veritas recommends a chunk size of 4k for SFCFSHA, where multiple copies of virtual machine images are accessed over NFS. For all other datasets, Veritas recommends a chunk size of 16k or higher.

The space consumed by the deduplication database is a function of the amount of data in the file system and the deduplication chunk size. The space consumed by the deduplication database grows with time as new data is added to file system. Additional storage is required for temporary use, such as sorting fingerprints. The temporary storage may be freed after the work completes. Ensure that sufficient free space is available for deduplication to complete successfully. The deduplication might not start if the file system free space is less than approximately 15%. The deduplication sometimes needs more than 15% free space for smaller chunk sizes. In general, the space consumed reduces significantly with larger chunk sizes. Veritas recommends that you have approximately 20% free space for 4k chunks.