| 1 | |
| 2 |
|
| 3 | Bigalloc
|
| 4 | --------
|
| 5 |
|
| 6 | At the moment, the default size of a block is 4KiB, which is a commonly
|
| 7 | supported page size on most MMU-capable hardware. This is fortunate, as
|
| 8 | ext4 code is not prepared to handle the case where the block size
|
| 9 | exceeds the page size. However, for a filesystem of mostly huge files,
|
| 10 | it is desirable to be able to allocate disk blocks in units of multiple
|
| 11 | blocks to reduce both fragmentation and metadata overhead. The
|
| 12 | bigalloc feature provides exactly this ability.
|
| 13 |
|
| 14 | The bigalloc feature (EXT4_FEATURE_RO_COMPAT_BIGALLOC) changes ext4 to
|
| 15 | use clustered allocation, so that each bit in the ext4 block allocation
|
| 16 | bitmap addresses a power of two number of blocks. For example, if the
|
| 17 | file system is mainly going to be storing large files in the 4-32
|
| 18 | megabyte range, it might make sense to set a cluster size of 1 megabyte.
|
| 19 | This means that each bit in the block allocation bitmap now addresses
|
| 20 | 256 4k blocks. This shrinks the total size of the block allocation
|
| 21 | bitmaps for a 2T file system from 64 megabytes to 256 kilobytes. It also
|
| 22 | means that a block group addresses 32 gigabytes instead of 128 megabytes,
|
| 23 | also shrinking the amount of file system overhead for metadata.
|
| 24 |
|
| 25 | The administrator can set a block cluster size at mkfs time (which is
|
| 26 | stored in the s_log_cluster_size field in the superblock); from then
|
| 27 | on, the block bitmaps track clusters, not individual blocks. This means
|
| 28 | that block groups can be several gigabytes in size (instead of just
|
| 29 | 128MiB); however, the minimum allocation unit becomes a cluster, not a
|
| 30 | block, even for directories. TaoBao had a patchset to extend the “use
|
| 31 | units of clusters instead of blocks” to the extent tree, though it is
|
| 32 | not clear where those patches went-- they eventually morphed into
|
| 33 | “extent tree v2” but that code has not landed as of May 2015.
|
| 34 |
|