Chapter 7. File System

XFS scalability

The XFS file system is currently supported in Red Hat Enterprise Linux 6 and is well suited for very large files and file systems on a single host. Integrated backup and restore, direct I/O and online resizing of the file system are some of the benefits that this file system provides.

The XFS implementation has been improved to better handle metadata intensive workloads. An example of this type of workload is accessing thousands of small files in a directory. Prior to this enhancement, metadata processing could cause a bottleneck and lead to degraded performance. To address this problem an option to delay the logging of the metadata has been added that provides a significant performance improvement. As a result of this delayed logging of metadata, XFS performance is on par with ext4 for such workloads. The default mount options have also been updated to use delayed logging.
Parallel NFS

Parallel NFS (pNFS) is a part of the NFS v4.1 standard that allows clients to access storage devices directly and in parallel. The pNFS architecture eliminates the scalability and performance issues associated with NFS servers in deployment today.

pNFS supports 3 different storage protocols or layouts: files, objects and blocks. The Red Hat Enterprise Linux 6.2 NFS client supports the files layout protocol.
To automatically enable the pNFS functionality, create the /etc/modprobe.d/dist-nfsv41.conf file with the following line and reboot the system:
alias nfs-layouttype4-1 nfs_layout_nfsv41_files
Now when the -o minorversion=1 mount option is specified, and the server is pNFS-enabled, the pNFS client code is automatically enabled.
This feature is a Technology Preview. For more information on pNFS, refer to http://www.pnfs.com/.
Asynchronous writes in CIFS

The CIFS (Common Internet File System) protocol allows for a unified way to access remote files on disparate operating systems. The CIFS client has traditionally only allowed for synchronous writes. This meant that the client process would not yield back control until the writes were successfully completed. This can lead to degraded performance for large transactions that take long to complete. The CIFS client has been updated to write data in parallel without the need to wait for the sequential writes. This change can now result in performance improvements up to 200%.

CIFS NTLMSSP authentication

Support for NTLMSSP authentication has been added to CIFS. In addition, CIFS now uses the kernel's crypto API.

autofs4 module

The autofs4 module has been updated to kernel version 2.6.38.

Fixed tracepoints for ext3 and jbd

Fixed tracepoints have been added to ext3 and jbd.

Mount options in superblock

Support for the -o nobarrier mount option in ext4, and its utilities: tune2fs, debugfs, libext2fs, has been added.