Tuesday, 21 June 2016

Understanding superblocks in Linux file systems

A disk consists of sectors - a block is made up of these sectors.

Available block sizes differ dependent on the filesystem - for example:

ext2: 1024 (1Kb), 2048 (2Kb) or 4096 (4Kb) bytes

ext3: 1024 (1Kb), 2048 (2Kb), 4096 (4Kb) or 8096 (8kB) bytes

ext4: 1024 (1Kb) to 65536 (64Kb) bytes

xfs: 512 (0.5kB) to 65536 (64kB) bytes

The block size is defined when initially creating the file system e.g.:

mkfs -t ext3 -b 4096 /dev/sda1

Performance Considerations

Typically if you are going to be storing large files on the disk you will want to ensure you have a large[er] block size - look at it this way: if you have a large file with many blocks (the blocks combine together to create the file) it will take longer to read all of those individual blocks than if you had larger blocks - meaning less blocks to read.

And in reverse - if you have a large amount of small files you will want a small block size as there will be a greater amount of blocks available to accommodate the files and because the blocks are smaller the system is having to read less sectors and hence improves the performance.

Hard disks

A hard disk will contain a (constant) sector size - usually 512. It is important to note that a block size can't be lower than that of the hard disks sector size  - it must be a multiple of it's sector size e.g. 1024,2048 and so on.

Block Groups 

(Specific to EXT file systems - see 'Allocation Groups' for XFS) Block groups are made of individual blocks - which makes reading and writing large amounts of data easier.

Each block group contains a redundant copy of the super block (mentioned below) and file system descriptors.

The physical structure of the file system is made up as follows:



Superblocks

A superblock contains important filesystem meta data like:

- Blocks per block group
- Number of free blocks in the filesystem
- Mount and write times
- File system state (e.g. clean or dirty mount)

Hence protecting superblocks is extremely important - and loss of a superblock can render a filesystem unmountable!


0 comments:

Post a comment