Uninterruptible Power Supplies (UPS)
It is recommended that a UPS be used for each power circuit or that a site UPS be used. If a site UPS
is used, then analysis should be done to insure that loss of one circuit does not cause the failure of
more than 50% of the nodes.
Multiple Power Distribution Units should be used. They should be connected to at least two different
power circuits. When a component (server, switch, etc) has two power cords, then they should be
connected to PDUs that are powered from different circuits.
If a component has more than two power cords, then they can be connected to different
PDU's that are all powered from separate power circuits. If this is not possible, then all
of the power cords of the component should be connected to the same PDU. In this case,
follow the guideline of ensuring that there is not power loss to 50% or more of the servers
in the cluster.
Power calculations for a PDU should be done with the assumption that another PDU or
power circuit in the cluster could be down. For example, if a server that has two power
supplies has one power cord connected to a PDU, then assume that all of the power for
the server needs to be supplied by that PDU. This accounts for PDU, power supply and
power circuit failures.
Power Distribution Testing
It is important to test a cluster after it is configured. Part of that test should include resilience to power
circuit, PDU, or UPS failure. Each of these components should be "failed" in turn to ensure that the cluster
will not fail.
MC/ServiceGuard for NAS 8000
HP expands its existing NAS offerings and continues to deliver industry leading storage solutions with the
release of its new NAS 8000 series of solutions designed to provide scalable, easy-to-manage,
heterogeneous NAS solutions for its VA and XP arrays, as well as NAS integration into its SAN solutions
for flexible file-sharing. These solutions offer a NAS-optimized OS for file serving and cluster technology -
from MC/ServiceGuard - for mission critical needs.
For more information regarding this solution, go to...
Disaster Tolerant Solution for Linux
HP's metropolitan cluster for Linux is an integrated solution, utilizing MC/ServiceGuard for Linux and the
Cluster Extension XP software (CLX) for the HP SureStore XP Disk Arrays.
CLX is host-based software that integrates Continuous Access (CA) XP remote mirroring with
MC/ServiceGuard failover/failback operations.
Integrating CLX with MC/ServiceGuard for Linux allows application packages to automatically fail over:
among local nodes that are attached to the same XP series array.
between one system that is attached locally to its XP frame and another remote node that is attached
locally to another XP frame.
The supported configuration has the following requirements: