Archived content

NOTE: this is an archived page and the content is likely to be out of date.

PRIMECLUSTER Introduction

- The Answer for Business Continuity -

Clustering PRIMEQUEST, SPARC Servers, and PRIMERGY servers using Fujitsu PRIMECLUSTER will dramatically improve system uptime. Even if one server stops all together, system continuity is maintained by the forcible switching of operations from the operational server onto its backup twin. When used with standard server virtualization software such as VMware and KVM, PRIMECLUSTER can also maximize server utilization by minimizing the portion of the server that needs recovery.

Fail-Safe Switching Even When Server Stops

Early detection and very fast switching

Early problem detection is the first lesson in maximizing system uptime.
By inter-working with the server hardware, PRIMECLUSTER can swiftly detect any failure and fast switch the applications to an alternative server.

All Fujitsu servers contain, as standard, dedicated and independent hardware monitoring processors. This means that even if a server hangs, the monitoring processors keep operating and assist the system administrator to diagnose the problem. In the meantime the necessary system recovery operation quickly continues in the background.
By inter-working with all these high availability mechanisms(1) PRIMECLUSTER can immediately detect the system hang and forcibly switch affected applications to the alternative server. (Figure 1. "PRIMECLUSTER switching mechanism")
(note 1)

PRIMEQUEST Management Board (MMB)
PRIMERGY integrated Remote Management Controller (iRMC)
SPARC M10 Extended System Control Facility (XSCF)
SPARC T7/M7/T5 Integrated Lights Out Manager (ILOM)

Figure 1. PRIMECLUSTER switching mechanism

Failure prevention during switching

To assure any server switchover is successful, PRIMECLUSTER regularly checks the disk and server hardware. This prevents unintended failure during switching.

High-Efficiency Cluster

PRIMECLUSTER can increase the server utilization within the cluster system.
PRIMECLUSTER can reduce switching time by only switching the minimum number of failed applications and their environment to the alternate node. When a failure occurs, this is much faster than switching the whole server node. The PRIMECLUSTER approach means customers do not need idle system resources that are only used following a failure.

High availability and efficiency

A mutual stand-by cluster is one method of achieving such cost-efficient benefits (Figure 2. “Mutual stand by cluster system”).

Under normal server status, applications run on both servers. Cluster application A runs on node 1, and cluster application B runs on node 2.
If application A fails, switching occurs. Then in node 2, cluster application A is transformed to active status.

Figure 2. A mutual stand-by cluster system

Clustering in a virtualized environment

In virtual environments using VMware or Red Hat Enterprise Linux, PRIMECLUSTER can minimize the portion that is switched.

If an application or a guest OS fails, its operation can be quickly resumed just by switching the corresponding Guest OS to another server. (Figure 3. “Guest OS switch triggered by application failure (Dual nodes) ” ) All other Guest OS, unaffected by the problem, remain where they are and continue to operate as before.

Another variant of this, possible with the mission critical PRIMEQUEST server, is single server cluster. Many failures are application failures rather than hardware failures. In this case PRIMECLUSTER can reestablish the failed guest OS as another virtual server on the PRIMEQUEST. See Figure. 4. “Guest OS switch triggered by application failure (Single node)”

Different from clustering between hardware partitions, such virtualized clustering does not require reservation of CPU or memory. This is where the cost-efficiency comes from.

Figure 3. Guest OS switch triggered by application failure (Dual nodes)

Figure 4. Guest OS switch triggered by application failure (Single node)

PRIMECLUSTER has the finest-grain fail over, which can genuinely minimize influences of application failures.
The only failed application can be switched if the relevant information is configured on PRIMECLUSTER.