Extent Based Allocation

Disk space is allocated in 1024-byte sectors to form logical blocks. VxFS supports logical block sizes of 1024, 2048, 4096, and 8192 bytes. The default block size depends on the size of the file system. For file systems up to 4 TB, the block size is 1K. The block size is 2K for file systems from 4 TB to 8 TB, 4K for file systems from 8 TB to 16 TB, and 8K for file systems larger than 16 TB (see “The VxFS Version 5 Disk Layout” for more information).

An extent is defined as one or more adjacent blocks of data within the file system. An extent is presented as an address-length pair, which identifies the starting block address and the length of the extent (in file system or logical blocks). VxFS allocates storage in groups of extents rather than a block at a time.

Extents allow disk I/O to take place in units of multiple blocks if storage is allocated in consecutive blocks. For sequential I/O, multiple block operations are considerably faster than block-at-a-time operations; almost all disk drives accept I/O operations of multiple blocks.

Extent allocation only slightly alters the interpretation of addressed blocks from the inode structure compared to block based inodes. A VxFS inode references 10 direct extents, each of which are pairs of starting block addresses and lengths in blocks. The VxFS inode also points to two indirect address extents, which contain the addresses of other extents:

Each indirect address extent is 8K long and contains 2048 entries. All indirect data extents for a file must be the same size; this size is set when the first indirect data extent is allocated and stored in the inode. Directory inodes always use an 8K indirect data extent size. By default, regular file inodes also use an 8K indirect data extent size that can be altered with vxtunefs (see “Tuning VxFS I/O Parameters” ); these inodes allocate the indirect data extents in clusters to simulate larger extents.

Typed Extents

In the Version 4 disk layout, VxFS introduced a new inode block map organization for indirect extents known as typed extents. Each entry in the block map has a typed descriptor record containing a type, offset, starting block, and number of blocks.

Indirect and data extents use this format to identify logical file offsets and physical disk locations of any given extent. The extent descriptor fields are defined as follows:

type	Uniquely identifies an extent descriptor record and defines the record's length and format.
offset	Represents the logical file offset in blocks for a given descriptor. Used to optimize lookups and eliminate hole descriptor entries.
starting block	The starting file system block of the extent.
number of blocks	The number of contiguous blocks in the extent.

• Indirect address blocks are fully typed and may have variable lengths up to a maximum and optimum size of 8K. On a fragmented file system, indirect extents may be smaller than 8K depending on space availability. VxFS always tries to obtain 8K indirect extents but resorts to smaller indirects if necessary.

• Indirect Data extents are variable in size to allow files to allocate large, contiguous extents and take full advantage of VxFS's optimized I/O.

• Holes in sparse files require no storage and are eliminated by typed records. A hole is determined by adding the offset and length of a descriptor and comparing the result with the offset of the next record.

• While there are no limits on the levels of indirection, lower levels are expected in this format since data extents have variable lengths.

• This format uses a type indicator that determines its record format and content and accommodates new requirements and functionality for future types.

The current typed format is used on regular files only when indirection is needed. Typed records are longer than the previous format and require less direct entries in the inode. Newly created files start out using the old format which allows for ten direct extents in the inode. The inode's block map is converted to the typed format when indirection is needed to offer the advantages of both formats.