GPFS: Administration and Programming Reference - IRA Home
GPFS: Administration and Programming Reference - IRA Home
GPFS: Administration and Programming Reference - IRA Home
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
Glossary<br />
This glossary defines technical terms <strong>and</strong><br />
abbreviations used in <strong>GPFS</strong> documentation. If you<br />
do not find the term you are looking for, refer to<br />
the index of the appropriate book or view the IBM<br />
Glossary of Computing Terms, located on the<br />
Internet at: w3.ibm.com/st<strong>and</strong>ards/terminology.<br />
B<br />
backup NSD server. A node designated to perform<br />
NSD disk access functions in the event that the primary<br />
NSD server fails.<br />
block utilization. The measurement of the percentage<br />
of used subblocks per allocated blocks.<br />
C<br />
cluster. A loosely-coupled collection of independent<br />
systems (nodes) organized into a network for the<br />
purpose of sharing resources <strong>and</strong> communicating with<br />
each other. (See also “<strong>GPFS</strong> cluster” on page 381).<br />
cluster configuration data files. The configuration<br />
data that is stored on the cluster configuration servers.<br />
configuration manager. The node that selects file<br />
system managers <strong>and</strong> determines whether quorum<br />
exists. The oldest continuously operating node in the file<br />
system group is automatically assigned as the<br />
configuration manager.<br />
control data structures. Data structures needed to<br />
manage file data <strong>and</strong> metadata cached in memory.<br />
Control data structures include hash tables <strong>and</strong> link<br />
pointers for finding cached data; lock states <strong>and</strong> tokens<br />
to implement distributed locking; <strong>and</strong> various flags <strong>and</strong><br />
sequence numbers to keep track of updates to the<br />
cached data.<br />
D<br />
Data Management Application Program Interface<br />
(DMAPI). The interface defined by the Open Group’s<br />
XDSM st<strong>and</strong>ard as described in the publication System<br />
Management: Data Storage Management (XDSM) API<br />
Common Application Environment (CAE) Specification<br />
C429, The Open Group ISBN 1-85912-190-X.<br />
deadman switch timer. A kernel timer that ensures<br />
that a node that has lost its disk lease <strong>and</strong> has<br />
outst<strong>and</strong>ing I/O requests cannot complete those<br />
requests (<strong>and</strong> risk causing file system corruption). This<br />
is achieved by panicking the kernel.<br />
disk descriptor. A disk descriptor defines how a disk<br />
is to be used within a file system. Each descriptor<br />
supplied to the mmcrfs comm<strong>and</strong> must be in the form<br />
(second, third <strong>and</strong> sixth fields reserved):<br />
DiskName:::DiskUsage:FailureGroup::StoragePool<br />
Each descriptor supplied to the mmcrnsd comm<strong>and</strong><br />
must be in the form:<br />
DiskName:PrimaryServer:BackupServer:DiskUsage:<br />
FailureGroup:DesiredName:StoragePool<br />
DiskName<br />
The block device name appearing in /dev for<br />
the disk you want to define as an NSD.<br />
Examples of disks accessible through a block<br />
device are SAN-attached disks or virtual<br />
shared disks. If a PrimaryServer node is<br />
specified, DiskName must be the /dev name<br />
for the disk device on the primary NSD server<br />
node. Use the <strong>GPFS</strong> FAQ link at<br />
http://publib.boulder.ibm.com/infocenter/clresctr/<br />
index.jsp?topic=/com.ibm.cluster.infocenter.doc/<br />
library.html for the latest supported disk types.<br />
Note that <strong>GPFS</strong> provides some helper<br />
comm<strong>and</strong>s to ease configuration of /dev disk<br />
devices. For instance the mmcrvsd comm<strong>and</strong><br />
can configure a virtual shared disk <strong>and</strong> make it<br />
accessible to nodes connected over a high<br />
performance switch. The output disk descriptor<br />
file from an mmcrvsd comm<strong>and</strong> can be used<br />
as input to the mmcrnsd comm<strong>and</strong> since the<br />
virtual shared disk names enumerated in that<br />
file will appear as /dev block devices on switch<br />
attached nodes.<br />
PrimaryServer<br />
The name of the primary NSD server node.<br />
If this field is omitted, the disk is assumed to<br />
be directly attached to all nodes in the cluster.<br />
BackupServer<br />
The name of the backup NSD server node.<br />
If the PrimaryServer has been specified <strong>and</strong><br />
this field is omitted, it is assumed you do not<br />
want failover in the event the PrimaryServer<br />
fails. If a PrimaryServer has not been specified,<br />
this field must also be omitted.<br />
Disk Usage<br />
What is to be stored on the disk:<br />
dataAndMetadata<br />
Indicates that the disk contains both<br />
data <strong>and</strong> metadata. This is the default.<br />
dataOnly<br />
Indicates that the disk contains data<br />
<strong>and</strong> does not contain metadata.<br />
© Copyright IBM Corp. 1998, 2006 379