Faster and Smaller qcow2 Files With Subcluster-based Allocation

Explore the innovative subcluster-based allocation technique for qcow2 files in this informative conference talk. Delve into the structure of qcow2 files, understanding the challenges of copy-on-write operations and their impact on I/O performance and space utilization. Learn how decreasing cluster sizes affects metadata management and discover the intricacies of L1 and L2 tables. Examine the concept of subcluster allocation, its implementation details, and the benefits it brings in terms of reduced copy-on-write operations, improved I/O speed, and optimized space usage. Gain insights into the current implementation status and acknowledge the contributors to this advancement in qcow2 file technology.

Intro
The qcow2 file format
Structure of a qcow2 file
copy-on-write means more I/O
copy-on-write means more used space
Decreasing the cluster size
Smaller clusters mean more metadata
The L1 table
L2 metadata size
Accessing L2 metadata
Reference counts
The overhead of having to allocate clusters
The situation so far
Subcluster allocation: what it looks like
L2 tables in details
Extended L2 entries
Two use cases for subcluster allocation
less copy-on-write means faster I/O
less copy-on-write means less used space
larger clusters mean less metadata
Implementation status
Acknowledgments