Startseite
Bild

napp-it-zfs

Parts if ZFS

If you want to understand ZFS, you should know about the components of ZFS,
the correct naming and the associated properties.


Pool


All data is stored in a pool.  A pool is build from one or more vdevs (virtual devices).  The size of a pool can grow practically without limits by adding vdevs. The often used default name "tank" for a pool is a good synonym for its usage.

Pool options:

  • Single vdev (Pool is build from one vdev). This is similar to a conventional Raid-Array.
    example: 11 disks that are organized as Raid-Z3. Best for storage up to 48TB (11 x 6TB) and Backup/Home/Media usage.

    or other example: Single 3way mirror - best for a capacity up to 6 TB (3 x 6TB) with best data security
  • Multiple Vdevs
    For high capacity storage up to Petabytes or when you need high IOPS values you can add more vdevs to a pool.
    All data is striped over all vdevs in a Raid-0 manner to improve performance. IOPS (Performance with concurrent random read/writes) scales therefor with the number of vdevs
  • Hotspare disks are defined on a pool level and replace a failed disk automatically
  • SSDs can be added to a pool and used as L2ARC read cache to extent the RAM based ARC readcache
  • DRAM or SSD disks can be added to a pool as a ZIL logdevice to increase performance of secure sync write. A ZIL is only needed and used when sync write is enabled.


vdev


You build a pool from one or more vdevs (virtual devices). A vdev is a single file or disk or build from multiple files or disks that are organized as a Raid. A Vdev cannot grow  in size by adding more disks but you can replace all disks by larger ones to increase vdev size.

Vdev options

  • Basic (single File or disk). A pool with one or multiple basic vdevs acts like a Raid-0 array without redundancy!
  • Mirror (2way or n-way). ZFS reads from all disks simulaniously. Best for high IOPS usage and spindle based disks. With SSD only vdevs you should prefer Raid-Z due to usable capacity.
  • Raid-Z1 (one parity disk, similar to Raid-5 but without write-hole problem. With any Raid-Z there is no need for a disc controller with BBU/battery). Best for 3 disks. You should avoid with larger or more disks due to rebuild time. (A loss of two disks = total loss of pool).
  • Raid Z2 (two parity disks, similar Raid-6 but without write hole problem). Best number of disks per vdev is 4, 6 or 10 disks. Avoid more disks due to rebuild time. (A loss of three disks = total loss of pool).
  • Raid Z3 (three parity disks). Best number of disks per vdev is 5, 7 or 11 disks. Avoid more disks due to rebuild time. (A loss of three disks = total loss of pool). 19 disks for a single vdev pool may be acceptable for some backups systems.


Filesystem


You can create filesystems on a pool. They are similar to conventional disk partitions but without a fixed size. They can grow up to poolsize dynamically. Usage is limited and managed by reservations and quotas - this is called storage virtualisation. Filesystems are mounted as folders and you can share them via NFS or CIFS or other protocols. You can create thousands of filesystems on a pool without problem.


Zvol


A zvol is a resizable blockdevice (a virtual disk). You can create zvols on a filesystem and use it like a disk example as a Comstar target for FC/ IB/ iSCSI.

napp-it 31.03.2017