In computing, ZFS is a file system originally created by Sun Microsystems for the Solaris Operating System (not to be mixed up with zFS, a file system found in IBM's z/OS). The features of ZFS include high storage capacity, integration of the concepts of filesystem and volume management, a novel on-disk structure, lightweight instances, and easy storage pool management. ZFS is implemented as open-source software, licensed under the Common Development and Distribution License (CDDL).

History

ZFS was designed and implemented by a team at Sun led by Jeff Bonwick. It was announced on September 14, 2004. Source code for ZFS was integrated into the main trunk of Solaris development on October 31, 2005 and released as part of build 27 of OpenSolaris on November 16, 2005. Sun announced that ZFS was included in the 6/06 update to Solaris 10 in June 2006, one year after the opening of the OpenSolaris community.

The name originally stood for "Zettabyte File System", but is now a pseudo-initialism.

Storage pools

Unlike traditional file systems, which reside on single devices andthereforerequire a volume manager to use more than one device, ZFS filesystems are built on top of virtual storage pools called zpools. A zpool is constructed of virtual devices (vdevs), which are themselves constructed of block devices: files, hard drive partitions, or entire drives, with the last being the recommended usage. Block devices within a vdev may be configured in different ways, depending on needs and space available: non-redundantly (similar to RAID 0), as a mirror (RAID 1) of two or more devices, as a RAID-Z group of three or more devices, or as a RAID-Z2 group of five or more devices. The storage capacity of all vdevs is available to all of the file system instances in the zpool.

A quota can be set to limit the amount of space a file system instance can occupy, and a reservation can be set to guarantee that space will be available to a file system instance.

Capacity

ZFS is a 128-bit file system, so it can store 18 billion billion (18.4 x 1018) times more data than current 64-bit systems. The boundaries of ZFS are designed to be so large that they will not be encountered in practice for some time. Some theoretical limits in ZFS are:

264 - Number of snapshots of any file system

248 - Number of entries in any individual directory

16 EiB (264 bytes) - Maximum size of a file system

16 EiB - Maximum size of a single file

16 EiB - Maximum size of any attribute

256 ZiB (278 bytes) - Maximum size of any zpool

256 - Number of attributes of a file (actually constrained to 248 for the number of files in a ZFS file system)

256 - Number of files in a directory (actually constrained to 248 for the number of files in a ZFS file system)

264 - Number of devices in any zpool

264 - Number of zpools in a system

264 - Number of file systems in a zpool

If a billion computers each filled a billion individual file systems per second, the time required to reach the limit of the overall system would be almost 1,000 times the estimated age of the universe.

Project leader Bonwick said, "Populating 128-bit file systems would exceed the quantum limits of earth-based storage. You couldn't fill a 128-bit storage pool without boiling the oceans." Later he clarified:

Copy-on-write transactional model

ZFS uses a copy-on-write transactional object model. All block pointers within the filesystem contain a 256-bit checksum of the target block which is verified when the block is read. Blocks containing active data are never overwritten in place; instead, a new block is allocated, modified data is written to it, and then any metadata blocks referencing it are similarly read, reallocated, and written. To reduce the overhead of this process, multiple updates are grouped into transaction groups, and an intent log is used when synchronous write semantics are required.

Snapshots and clones

An advantage of copy-on-write is that when ZFS writes new data, the blocks containing the old data can be retained, allowing a snapshot version of the file system to be maintained. ZFS snapshots are created very quickly, since all the data composing the snapshot is already stored; they are also space efficient, since any unchanged data is shared among the file system and its snapshots.

Writeable snapshots ("clones") can also be created, resulting in two independent file systems that share a set of blocks. As changes are made to any of the clone file systems, new data blocks are created to reflect those changes, but any unchanged blocks continue to be shared, no matter how many clones exist.

Dynamic striping

Dynamic striping across all devices to maximize throughput means that as extra devices are added to the zpool, the stripe width automatically expands to include them;thereforeall disks in a pool are used, which balances the write load across them.

Variable block sizes

ZFS uses variable-sized blocks of up to 128 kilobytes. The present ly available code allows the administrator to tune the maximum block size used as certain workloads do not perform well with large blocks. Automatic tuning to match workload characteristics is contemplated.

If data compression is enabled, variable block sizes are used. If a block can be compressed to fit into a smaller block size, the smaller size is used on the disk to use less storage and improve IO throughput (though at the cost of increased CPU use for the compression and decompression operations).

Lightweight filesystem creation

In ZFS, filesystem manipulation within a storage pool is easier than volume manipulation within a traditional filesystem; the time and effort required to create or resize a ZFS filesystem is closer to that of making a new directory than it is to volume manipulation in some other systems.

Additional capabilities

Explicit I/O priority with deadline scheduling.

Claimed globally optimal I/O sorting and aggregation.

Multiple independent prefetch streams with automatic length and stride detection.

Parallel, constant-time directory operations.

End-to-end checksumming, allowing data corruption detection (and recovery if you have redundancy in the pool).

Intelligent scrubbing and resilvering.

Load and space usage sharing between disks in the pool.

Ditto blocks: Metadata is replicated inside the pool, two or three times (according to metadata importance). If the pool has several devices, ZFS tries to replicate over different devices. So a pool without redundancy can lose data if you find bad sectors, but metadata should be rather safe even in this scenario.

ZFS design (copy-on-write + superblocks) is safe when using disks with write cache enabled, if they support the cache flush commands issued by ZFS. This feature provides safety and a performance boost compared with some other filesystems.

When entire disks are added to a ZFS pool, ZFS automatically enables their write cache. This is not done when ZFS only manages discrete slices of the disk, since it doesn't know if other slices are managed by non-write-cache safe filesystems, like UFS.

Filesystem encryption is supported, though is currently in an alpha stage.

Cache management

ZFS also uses the ARC, a new method for cache management, instead of the traditional Solaris virtual memory page cache.

Limitations

ZFS does not support per-user or per-group quotas. Instead, it is possible to create user-owned filesystems, each with its own size limit. Intrinsically, there is no practical quota solution for the file systems shared among several users (such as team projects, for example), where the data can't be separated per user, although it could be implemented on top of the ZFS stack.

Capacity expansion is normally achieved by adding groups of disks as vdev (stripe, RAID-Z, RAID-Z2, or mirrored). Newly written data will dynamically start to use all available vdevs. It is also possible to expand the array by iteratively swapping each drive in the array with a bigger drive and waiting for ZFS to heal itself - the heal time will depend on amount of stored information, not the disk size. If a snapshot is taken during this process, it will cause the heal to be restarted.

It is currently not possible to reduce the number of vdevs in a pool nor otherwise reduce pool capacity. However, it is currently being worked on by the ZFS team.

It is not possible to add a disk to a RAID-Z or RAID-Z2 vdev. This feature appears very difficult to implement.

Reconfiguring storage requires copying data offline, destroying the pool, and recreating the pool with the new policy.

ZFS is not a native cluster, distributed, or parallel file system and can't provide concurrent access from multiple hosts as ZFS is a local file system. The Lustre distributed filesystem is being adapted to use ZFS as back-end storage for both data and metadata.

It is currently not possible to natively nest vdevs within a zpool - to create, for example, a non-redundant pool of RAID-Z vdevs (similar to a RAID 5+0).

Solaris implementation issues

The present ZFS implementation (Solaris 10 11/06) has some issues administrators should know before deploying it. Many of these issues are scheduled to be addressed in future releases.

ZFS root filesystem support is currently set to off on Solaris 10 default installations, since the standard installer still does not fully support ZFS roots. The ZFS Boot project successfully added boot support to the OpenSolaris project in March 2007. Bootable ZFS file systems are only available for x86 systems, and must be done with scripts or manually. As of February 2008, Sun has said that supported ZFS boot for both SPARC and x86 systems is planned for a Solaris 10 update in late 2008.

If a Solaris Zone is put on ZFS, the system can't be upgraded; the OS will need to be reinstalled. This issue is planned to be addressed in a Solaris 10 update in 2007 .[update

A file "fsync" will commit to disk all pending modifications on the filesystem. That is, an "fsync" on a file will flush out all deferred (cached) operations to the filesystem (not the pool) in which the file is located. This can make some fsync() slow when running alongside a workload which writes a lot of data to filesystem cache. The issue is currently fixed in Solaris Nevada.

New vdevs can be added to a storage pool, but they can't be removed. A vdev can be exchanged for using a bigger new one, but it can't be removed, in the process reducing the total pool storage size even if the pool has enough unused space. The ability to shrink a zpool is a work in progress, currently targeted for a Solaris 10 update in late 2007.[update

ZFS encourages creation of many filesystems inside the pool (for example, for quota control), but importing a pool with thousands of filesystems is a slow operation (can take minutes).

ZFS uses a lot of CPU when doing small writes (for example, a single byte). There are two root causes, currently being worked on: a) Translating from znode to dnode is slower than necessary because ZFS doesn't use translation information it already has, and b) Current partial-block update code is very inefficient.

ZFS copy-on-write operation can degrade on-disk file layout (file fragmentation) when files are modified, decreasing performance.

ZFS blocksize is configurable per filesystem, currently 128 KB by default. Reads or writes which are smaller than the block size suffer a performance penalty. If your workload reads/writes data in fixed sizes (blocks), for example a database, you should (manually) configure ZFS blocksize equal to the application blocksize, for better performance and to conserve cache memory and disk bandwidth.

ZFS only offlines a faulty hard disk if it can't be opened. Read/write errors or slow/timed-out operations do not currently cause a disk to be marked as faulty. This is fixed in Solaris Nevada via 6520519.

When listing ZFS space usage, the "used" column only shows non-shared usage; if some data is shared (for example, between snapshots), it can't be determined how much unique data is held by a specific snapshot, making it hard to tell, for example, which set of snapshots must be destroyed to free up space.

There is work in progress to give automatic and periodic disk scrubbing, in order to give corruption detection and early disk-rotting detection. Currently the data scrubbing must be done manually with "zpool scrub" command.

Current ZFS compression/decompression code is very fast, but the compression ratio is not comparable to gzip or similar algorithms. The gzip compression algorithm was added in Solaris Nevada as part of 6536606 and is planned for a Solaris 10 update in Spring 2008.

If a snapshot is taken or destroyed while the zpool is scrubbing/resilvering, the process will be restarted from the beginning.

Not all symbolic links are protected by ditto blocks.

Swapping over ZVOL pseudo-devices can hang the system.

Platforms

ZFS is part of Sun's own Solaris operating system and isthereforeavailable on both SPARC and x86-based systems. Since the code for ZFS is open source, a port to other operating systems and platforms can be produced without Sun's involvement.

Nexenta OS

Nexenta OS, a complete GNU-based open source operating system built on top of the OpenSolaris kernel and runtime, includes a ZFS implementation, added in version alpha1. More recently, Nexenta Systems announced NexentaStor, their ZFS storage appliance based on Nexenta OS. NexentaStor includes a GUI that simplifies the process of utilizing ZFS.

Mac OS X

In a post on the opensolaris.org zfs-discuss mailing list, Apple Inc. announced it was porting ZFS to their Mac OS X operating system. From Mac OS X 10.5 Developer Seed 9A321, support for ZFS has been included, but lacks the ability to act as a root partition, noted above. Also, attempts to format local drives using ZFS were unsuccessful; this is a known bug.

On June 6, 2007, Sun's CEO Jonathan I. Schwartz announced that Apple would make ZFS "the" filesystem in Mac OS 10.5 Leopard. Marc Hamilton, VP for Solaris Marketing later wrote to clarify that, in his opinion, Apple is planning to use ZFS in future versions of Mac OS X, but not necessarily as the default filesystem for Mac OS X 10.5 Leopard. In the release version of Mac OS X 10.5, ZFS is available in read-only mode from the command line, which lacks the possibility to create zpools or write to them, but Apple has also released the "ZFS Read/Write Developer Preview 1.1 for Leopard", which allows read-write access and the creation of zpools. The installer for the "ZFS Read/Write Developer Preview 1.1 for Leopard" is currently only working on version 10.5.0, and has not been updated for the current version 10.5.1. As of January 2008, Apple provides read-write binaries and source, but they must be installed by hand.

Linux

Porting ZFS to Linux is complicated by the fact that the GNU General Public License, which governs the Linux kernel, prohibits linking with code under certain licenses, such as CDDL, the license ZFS is released under. One solution to this problem is to port ZFS to Linux's FUSE system so the filesystem runs in userspace instead. A project to do this was sponsored by Google's Summer of Code program in 2006, and is in Beta stage as of May 2007. Running a file system outside the kernel on traditional Unix-like systems can have a significant performance impact. However, NTFS-3G (another file system driver built on FUSE) performs well when compared to other traditional file system drivers. This shows that excellent performance is possible with ZFS on Linux after proper optimization. Sun Microsystems has stated that a Linux port is being investigated.

BSD

Pawel Jakub Dawidek has ported and committed ZFS to FreeBSD for inclusion in FreeBSD 7.0, due to be released in 2008.

As a part of the 2007 Google Summer of Code ZFS is being ported to NetBSD.

Other

There are no plans to port ZFS to HP-UX or AIX.

Adaptive Endianness

Pools and their associated ZFS file systems can be moved between different platform architectures, including systems implementing different byte orders. The ZFS block pointer format stores filesystem metadata in an endian-adaptive way; individual metadata blocks are written with the native byte order of the system writing the block. When reading, if the stored endianness doesn't match the endianness of the system, the metadata is byte-swapped in memory.

This does not affect the stored data itself; as is usual in POSIX systems, files appear to applications as simple arrays of bytes, so applications creating and reading data remain accountable for doing so in a way independent of the underlying system's endianness.