53
move to the other server; when the failed server comes back online, its services will automatically move back and normal operations resume as they were pre-failure.
This feature is not dependent upon services – you can have iSCSI targets split over both servers, or run iSCSI targets on one server, with CIFS/AFP services on the other. Resources created on the same cluster drive will run together on the same node.
The JMR SHARE with HA (High-Availability) Cluster is an ideal solution for IT professionals, as you gain not only high availability storage with continuous replication but also a very efficient and easy-to-use management interface. The server and cluster configuration can be completed in less than ten minutes.
Architecture
The high-availability archi-tecture includes the following:
• Cluster node: This is the individual server that is a member of the cluster; all nodes are equal, and from each node the user can monitor and configure cluster services.
• Cluster drive: This repre-sents a mirror of two drives on each individual server; the two drives can be a single disk, or preferably a RAID array, which are mirrored in real-time. On the drives, shares and iSCSI targets are installed. For example, each individual
server contains 16 disk drives (as in the JMR SHARE), providing up to 128TB capacity and each “drive” (RAID array) can be defined to run on the server. All resources defined on this cluster drive will run on the server; this way, the load between the servers will be balanced to achieve the best possible performance.
• Cluster resource: This represents network share or an iSCSI target within the cluster, and may be SMB/CIFS & AFP, NFS share or an iSCSI target. Each resource has its own size and is accessible through the IP address defined for the cluster drive.
• Corporate network: The network used by clients for accessing cluster resources.
• Replication network: Inter-nal network used for heartbeat and data replication between servers.
• Cluster access IP address: The address that enables clients to access shares and iSCSI targets. This address is shared by both HA cluster systems; if any server fails, the cluster IP address will move to the remaining server.
• Network test IP: This prevents storage desynchroni-zation (“split brain”) situations which might occur when both nodes are running but one or both are network dis-connected. HA Cluster nodes ping this address constantly and check if it is still able to reach the network. IP addresses on any pingable device on the network that is constantly powered (router, gateway, printer, mail server, etc.) can be used as network test IP. If one of the nodes is unable to ping this IP, it will automatically know it cannot reach the network and push all services over to the other node. If a node cannot reach all defined test IP addresses, it will shut down services to assure data remains consistent.
Cluster Events
Within the cluster, different types of events can occur. Depending upon the type of event, transfer of services can take varying amounts of time. Initial time for servers to realize the other server has failed is about 30 seconds; after this, resources are moved quickly.
Estimated time to move ten resources on another server is less than 60 seconds. To move 30 resources usually takes 70-80 seconds. This depends on the performance of the hardware and their storage capacity. The JMR SHARE is optimized to perform these transfers as expeditiously as possible even when compared with systems at substantially higher cost.
• Failover: This occurs when a server detects a fault and automatically moves all resources to the other server.
• Switchover: Switchover occurs when a resource is manually moved from one server to another. This may be performed when node preference settings are changed, or during system management of one of the nodes.