Go to main content

Oracle® ZFS Storage Appliance Administration Guide, Release OS8.7.x

Exit Print View

Updated: November 2018

Data Profiles for Storage Pools

After storage devices are physically verified and resources are allocated for a storage pool, the next step is to choose a storage profile that reflects your reliability, availability, serviceability (RAS), and performance goals. The set of possible profiles presented depends on your available storage. The following table lists all possible profiles and their descriptions.

Table 24  Data Profiles
Data Profile
Dual Parity Options
Triple mirrored
Data is triply mirrored, yielding a very highly reliable and high-performing system (for example, storage for a critical database). This configuration is intended for situations in which maximum performance and availability are required. Compared with a two-way mirror, a three-way mirror adds additional IOPS per stored block and higher level protection against failures. Note: A controller without expansion storage should not be configured with triple mirroring.
Double parity RAID
RAID in which each stripe contains two parity disks. As with triple mirroring, this yields high availability, as data remains available with the failure of any two disks. Double parity RAID is a higher capacity option than the mirroring options and is intended either for high-throughput sequential-access workloads (such as backup) or for storing large amounts of data with low random-read component.
Single Parity Options
Data is mirrored, reducing capacity by half, but yielding a highly reliable and high-performing system. Recommended when space is considered ample, but performance is at a premium (for example, database storage).
Single parity RAID, narrow stripes
RAID in which each stripe is kept to three data disks and a single parity disk. For situations in which single parity protection is acceptable, single parity RAID offers a much higher capacity option than simple mirroring. This higher capacity needs to be balanced against a lower random read capability than mirrored options. Single parity RAID can be considered for non-critical applications with a moderate random read component. For pure streaming workloads, give preference to the Double parity RAID option which has higher capacity and more throughput.
Data is striped across disks, with no redundancy. While this maximizes both performance and capacity, a single disk failure will result in data loss. This configuration is not recommended. For pure streaming workloads, consider using Double parity RAID. Due to non-redundancy, disks configured in a striped profile will not receive firmware updates, unless the configured storage pools are in an exported state.
Triple parity RAID, wide stripes
RAID in which each stripe has three disks for parity. This is the highest capacity option apart from Striped Data. Resilvering data after one or more drive failures can take significantly longer due to the wide stripes and low random I/O performance. As with other RAID configurations, the presence of cache can mitigate the effects on read performance. This configuration is not generally recommended.

Note -  Earlier software versions supported double parity with wide stripes. This has been supplanted by triple parity with wide stripes, as it adds significantly better reliability. Pools configured as double parity with wide stripes under a previous software version continue to be supported, but newly-configured or reconfigured pools cannot select that option.

NSPF Option

For expandable systems, some profiles may be available with an 'NSPF' option. This stands for 'no single point of failure' and indicates that data is arranged in mirrors or RAID stripes such that a pathological disk shelf failure will not result in data loss. Note that systems are already configured with redundancy across nearly all components. Each disk shelf has redundant paths, redundant controllers, and redundant power supplies and fans. The only failure that NSPF protects against is disk backplane failure (a mostly passive component), or gross administrative misconduct (detaching both paths to one disk shelf). In general, adopting NSPF will result in lower capacity, as it has more stringent requirements on stripe width.

Log Devices

Log devices can be configured using only striped or mirrored profiles. Log devices are only used in the event of node failure. For data to be lost with unmirrored logs, it is necessary for both the device to fail and the node to reboot immediately after. This a highly-unlikely event, however mirroring log devices can make this effectively impossible, requiring two simultaneous device failures and node failure within a very small time window.

Note -  When different sized log devices are in different chassis, only striped log profiles can be created.

Cache Devices

In a cluster configuration, cache devices installed in controller slots are available only to the controller which has the storage pool imported. In a cluster, it is possible to configure cache devices on both controllers to be part of the same pool. To do this, take over the pool on the passive node, then add storage, and select the cache devices. This has the effect of having half the global cache devices configured at any one time. While the data on the cache devices will be lost on failover, the new cache devices can be used on the new controller.

Cache devices installed in disk shelf slots, when added to a pool, are automatically imported during a cluster failback or takeover. No additional pool configuration is required.

Meta Devices

A meta device is a cache device used to store deduplicated metadata and other metadata for projects and shares. Meta devices can be allocated to a storage pool, but not an all-flash storage pool, during and after storage pool creation. However, they cannot be re-configured as normal cache devices for a pool, nor can they be removed from a pool. A meta device must be a 3.2 TB (minimum) SSD to support the enhanced data deduplication feature available in software version OS8.7.0 (2013.1.7.0) or later.

Before using meta devices and the deduplication feature for new and existing storage pools, accept the deferred software update for Data Deduplication v2, introduced with software version OS8.7.0 (2013.1.7.0). If replicating to other systems, both the replication source and targets must have this deferred update. For more information, see Data Deduplication, and Data Deduplication v2 Deferred Update in Oracle ZFS Storage Appliance Customer Service Manual.

Hot Spares

Hot spares are allocated as a percentage of total pool size and are independent of the profile chosen (with the exception of striped, which doesn't support hot spares). Because hot spares are allocated for each storage configuration step, it is much more efficient to configure storage as a whole than it is to add storage in small increments.

Related Topics:

  • Creating a Storage Pool (BUI, CLI).

  • Adding a Cache, Meta, or Log Device to an Existing Storage Pool (BUI, CLI).