zfs.txt

disk connection type and protocol:
1. SCSI (Small Computer System Interface), iSCSI
2. FC (Fiber Channel), FCoE (Fiber Channel over Ethernet)
3. IDE (Integrated Drive Electronics), ATA (Advanced Technology Attachment),
SATA (Serial ATA)

feature: 
a general-purpose POSIX file system using pool storage, transactional
copy-on-write of all blocks, an object-based storage model, and self-validating
checksums

pooled storage:
1. manage storage in the similar way as virtual memory: decouple file system 
from physical storage, multiple file systems should be able to share one pool 
of storage, rather than one-to-one association between file system and a 
particular storage device (or potion thereof)
2. implement feature of volume manager (i.e. mirroring) within ZFS

storage model (3.1):
traditional storage stack:
(1) device driver exports block device interface to volume manager
(2) VM exports another block device interface to file system
(3) file system exports vnode operation to syscall layer

ZFS storage stack:
(1) device driver exports block device interface to SPA (Storage Pool Allocator)
(2) SPA handles block allocation and IO, exports virtually addressed, explicitly
allocated and freed blocks to DMU (Data Management Unit)
(3) DMU turns virtually addressed blocked into a transactional object interface
for ZPL (ZFS POSIX layer)
(4) ZPL implements POSIX file system on top of DMU objects, and exports vnode
operations to syscall layer

technical details:
1. (checksum) checksum is kept at parent block, to make simutaneous corruption
of data and checksum less likely
2. on-disk consistency is maintained by DMU, rather than at the block allocation
level
3. (SPA) allocate in the unit of block
NOTE: LFS's allocation unit from small to big: block, extent, segment
(1) segment cleaner could introduce large overhead
(2) slab allocator could meet the demand for large continuous space
4. (DMU) implements transactional object, identified by 64-bit ID
(0) blocks are organized in tree structure
(1) all blocks are written in copy-on-write
(2) update for uberblock (root of block tree) is atomic in-place update