Administrator Guide

Data Caching and Redundancy
New and modified files are first written to the cache, and then cache data is immediately mirrored to the peer NAS controller (mirroring
mode). Data caching provides high performance, while cache mirroring between peer NAS controllers ensures data redundancy. Cache
data is ultimately transferred to permanent storage asynchronously through optimized data-placement schemes.
When cache mirroring is not possible, such as a single NAS controller failure or when the BPS battery status is low, NAS controllers write
directly to storage (journaling mode).
File Metadata Protection
The FluidFS cluster has several built-in measures to store and protect file metadata (which includes information such as name, owner,
permissions, date created, date modified, and a soft link to the file’s storage location).
• All metadata updates are recorded constantly to storage to avoid potential corruption or data loss in the event of a power failure.
• Metadata is replicated on two separate volumes.
• Metadata is managed through a separate caching scheme.
• Checksums protect the metadata and directory structure. A background process continuously checks and fixes incorrect checksums.
Load Balancing and High Availability
For availability and performance, client connections are load balanced across the available NAS controllers. Both NAS controllers in a NAS
appliance operate simultaneously. If one NAS controller in a NAS appliance fails, clients fail over automatically to the peer controller. When
failover occurs, some SMB clients will automatically reconnect to the peer NAS controller. In other cases, an SMB application might fail
and you must restart it. NFS clients experience a temporary pause during failover, but client network traffic resumes automatically.
Failure Scenarios
The FluidFS cluster can tolerate a single NAS controller failure without impact to data availability and without data loss. If one NAS
controller in a NAS appliance becomes unavailable (for example, because the NAS controller failed, is turned off, or is disconnected from
the network), the NAS appliance status is degraded. Although the FluidFS cluster is still operational and data is available to clients, you
cannot perform most configuration modifications, and performance might decrease because data is no longer cached.
The impact to data availability and data integrity of a multiple NAS controller failure depends on the circumstances of the failure scenario.
Detach a failed NAS controller as soon as possible, so that it can be safely taken offline for service. Data access remains intact as long as
one of the NAS controllers in each NAS appliance in a FluidFS cluster is functional.
The following table summarizes the impact to data availability and data integrity of various failure scenarios.
Scenario
System Status Data Integrity Comments
Single NAS controller failure Available, degraded Unaffected
• Peer NAS controller enters journaling mode
• Failed NAS controller can be replaced while
keeping the file system online
Sequential dual‐NAS controller
failure in single NAS appliance
cluster
Unavailable Unaffected Sequential failure assumes enough time is
available between NAS controller failures to write
all data from the cache to disk (Storage Center or
nonvolatile internal storage)
Simultaneous dual-NAS
controller failure in single NAS
appliance cluster
Unavailable Lose data in cache Data that has not been written to disk is lost
Sequential dual‐NAS controller
failure in multiple NAS appliance
cluster, same NAS appliance
Unavailable Unaffected Sequential failure assumes enough time is
available between NAS controller failures to write
all data from the cache to disk (Storage Center or
nonvolatile internal storage)
Simultaneous dual‐NAS
controller failure in multiple NAS
appliance cluster, same NAS
appliance
Unavailable Lose data in cache Data that has not been written to disk is lost
328 FluidFS Administration