Hardware Reference
In-Depth Information
the locking strategy accordingly. For directories that are only accessed by a
single node, it grants a token that covers the whole directory, so subsequent
operations within that directory incur no additional overhead for distributed
locking.
9.2.4 Advanced Data Management
GPFS supports a wide range of standard and advanced file system features;
three of these are described in more detail below. Other notable features in-
clude access control lists; extended attributes; immutability and append-only
restrictions; quotas; snapshots of all or part of a file system, including individ-
ual files (clones); defragmentation; rebalancing; and online reconfiguration.
9.2.4.1
GPFS Native RAID
In addition to support for traditional RAID controllers, GPFS offers an ad-
vanced software RAID implementation integrated into the NSD server called
GPFS Native RAID (GNR) [7, 8]). Using a conventional, dual-ported disk
enclosure lled with disks in a JBOD (or \just a bunch of disks") congu-
ration, GNR implements sophisticated data placement and error correction
algorithms to deliver high levels of storage reliability, availability, and perfor-
mance. GNR offers a variety of data protection techniques ranging from simple
replication up to three fault-tolerant Reed{Solomon codes. It declusters data,
parity and spare space across large numbers of disks to speed up rebuild times
and minimize impact on the foreground workload during rebuild. Write ver-
sion number tracking and end-to-end checksums allow it to detect and recover
from silent disk errors, such as dropped writes or off-track writes, as well as
network transport errors. A background scrubbing process verifies data/par-
ity consistency and data checksums to detect and fix silent disk corruption or
latent sector errors before additional errors might render them uncorrectable.
GNR can also exploit NVRAM and twin-tailed SSD as a multi-level fast write
cache for its own internal metadata updates and to speed up small writes.
9.2.4.2
Information Lifecycle Management
Applications have diverse reliability and performance demands of the
datasets they access, and the storage requirements for a particular piece
of data frequently changes over the lifetime of the data. To accommodate
such applications eciently and economically, GPFS supports powerful policy-
driven automated storage management through storage pools, filesets and user-
defined policies that match data to its most appropriate and cost-effective type
of storage [8].
Storage pools allow partitioning a file system into collections of storage
devices with similar properties that are managed together as a group. In
addition to the system pool, which contains all metadata and possibly some of
its data, a file system may have one or more data pools, which store file data
