Veritas File System 5.1 SP1 Administrator"s Guide (5900-1499, April 2011)

Partitioned directories
Normally, a large volume of parallel threads performing access and updates on a
directory that commonly exist in an file system suffers from exponentially longer
wait times for the threads. This feature creates partitioned directories to improve
the directory performance of file systems. When any directory crosses the tunable
threshold, this feature takes an exclusive lock on the directory inode and
redistributes the entries into various respective hash directories. These hash
directories are not visible in the name-space view of the user or operating system.
For every new create, delete, or lookup thread, this feature performs a lookup for
the respective hashed directory (depending on the target name) and performs the
operation in that directory. This leaves the parent directory inode and its other
hash directories unobstructed for access, which vastly improves file system
performance.
This feature operates only on disk layout Version 8 or later file systems.
Veritas File System performance enhancements
Traditional file systems employ block-based allocation schemes that provide
adequate random access and latency for small files, but which limit throughput
for larger files. As a result, they are less than optimal for commercial
environments.
VxFS addresses this file system performance issue through an alternative
allocation method and increased user control over allocation, I/O, and caching
policies.
See Using Veritas File System on page 30.
VxFS provides the following performance enhancements:
Data synchronous I/O
Direct I/O and discovered direct I/O
Support for file systems up to 256 terabytes
Support for files up to 16 terabytes
Enhanced I/O performance
Caching advisories
Enhanced directory features
Explicit file alignment, extent size, and preallocation controls
Tunable I/O parameters
Tunable indirect data extent size
Introducing Veritas File System
Veritas File System performance enhancements
28