ExpressCluster X 3.1 for Linux Getting Started Guide - Nec

High Availability (HA) clusterShared disk typeData must be inherited from one server to another in cluster systems. A cluster topology where datais stored in a shared disk with two or more servers using the data is called shared disk type.Shared Disk TypeData Mirror TypeMirroring- Expensive since a shared disk isnecessary.- Ideal for the system that handleslarge data- Cheap since a shared disk isunnecessary.- Ideal for the system with less datavolume because of mirroring.Figure 1-1: HA cluster configurationIf a failure occurs on a server where applications are running (active server), the cluster systemdetects the failure and applications are automatically started in a stand-by server to take overoperations. This mechanism is called failover. Operations to be inherited in the cluster systemconsist of resources including disk, IP address and application.In a non-clustered system, a client needs to access a different IP address if an application isrestarted on a server other than the server where the application was originally running. In contrast,many cluster systems allocate a virtual IP address on an operational basis. A server where theoperation is running, be it an active or a stand-by server, remains transparent to a client. Theoperation is continued as if it has been running on the same server.File system consistency must be checked to inherit data. A check command (for example, fsck orchkdsk in Linux) is generally run to check file system consistency. However, the larger the filesystem is, the more time spent for checking. While checking is in process, operations are stopped.For this problem, journaling file system is introduced to reduce the time required for failover.Logic of the data to be inherited must be checked for applications. For example, roll-back orroll-forward is necessary for databases. With these actions, a client can continue operation only byre-executing the SQL statement that has not been committed yet.A server with the failure can return to the cluster system as a stand-by server if it is physicallyseparated from the system, fixed, and then succeeds to connect the system. Such returning isacceptable in production environments where continuity of operations is important.Section I Introducing ExpressCluster17