Server cluster Technology in-depth understanding of the secret

In the early stages of development, a single processor can power a server and all its applications. Then it developed into a multiprocessing era, when two or more processors shared a single storage pool and were able to handle more and larger applications. Then a server network appears, each server in the network specializes in different application sets. Now, with the server cluster, two or more servers work like a server, delivering higher availability and performance, far beyond your imagination. Applications can be moved from one server to another, or simultaneously on several servers-all of which are transparent to the user.
&http://www.aliyun.com/zixun/aggregation/37954.html ">nbsp; Clusters are not new, but they are proprietary in software and hardware until recently. Information systems managers have taken the cluster more carefully because they can now use the standard hardware of mass production to implement clusters such as RAID, symmetric multi-processing system, network and I/O network cards and peripherals. Cluster technology in the future will be more development, now, the introduction of new cluster options, and the real cluster standards are still in place.

What is a cluster?

Simply put, clusters are two or more computers or nodes working together within a group. Clusters can provide higher availability and scalability than computers that work alone. Each node in a cluster usually has its own resources (processor, I/O, memory, operating system, storage) and is responsible for its own user set.

Failover function provides the silk to take the Mi yu the dry currency Huan Bibroni The Jake ⑸, to receive the fresh security the test Watts 芄 quot; switch to one or more other nodes in the cluster. Once the failed node resumes full operation, by proactively "switching" the functionality of one server to other servers in the cluster, you can upgrade, stop the server from running to increase the component, then put it back in the cluster, and then return its functionality from other servers to the server. With distributed message Delivery (DMP), which provides additional scalability, DMP is a cluster-based communication technology that allows applications to be extended to a single symmetric, multiple-processing (SMP) system in a transparent manner to end users.

Each node in the cluster must run cluster software to provide services such as fault detection, recovery, and the ability to manage the server as an approximate system. Nodes in a cluster must be connected in a way that knows all other node states. This is usually done through a communication path that is separated by a LAN path, and uses a dedicated NIC to ensure clear communication between nodes. The communication path relays a quot heartbeat to the system so that if a resource fails and cannot send a heartbeat, the failover process begins. In fact, the most reliable configuration uses a redundant heartbeat using different communication connections (LAN, SCSI, and RS232) to ensure that communication failures do not activate faulty failover.

Cluster level

Today, for cluster buyers, it's fortunate to have a variety of different classes of clusters to choose from, and they offer a wide range of usability. Of course, the higher the availability, the higher the price, and the greater the management complexity.

Shared storage

A shared disk subsystem is often the basis of a cluster, using shared SCSI or Fibre Channel. Each node uses its local disk storage operating system to swap space and system files, and the application data is stored on the shared disk, and each node can read data written by other nodes. Concurrent disk access between applications requires a Distribution lock manager (DLM), and the distance between the shared disk subsystem and its cluster nodes is limited by the selected media (SCSI or Fibre Channel, etc.).

Server mirroring (Mirrored disk)

Environments that require data redundancy without consuming additional disk subsystems have permission to select mirrored data between servers. In addition to the lower cost, another advantage of server mirroring is that the connection between the motherboard server and the secondary server can be lan-based, eliminating the SCSI distance limit. After the data is written to the motherboard server, it is also written to the secondary server, maintaining the integrity of the data by locking the server data. Some server mirroring products can also convert workloads from the primary server to the secondary server.

unshared

Some cluster products now use a "unshared" architecture in which the nodes neither share a centralized disk nor mirror data between nodes. In the event of a failure, a nonshared cluster has software that can transfer disk ownership from one node to another without using the distributed Distributed Lock Manager (DLM).

How do I make failover?

You can use a variety of methods to configure a cluster for failover. The first is the N-way configuration, where all nodes in the cluster normally have their own users and workloads. The resources of one failed node can be switched to other nodes, but their performance will decrease because the remaining servers assume additional load.

The n configuration includes a hot standby system that is in idle mode until the main system fails. In n configuration, when a node fails, the performance of other nodes can be avoided. However, because the standby node does not normally provide services, the cost is higher.

In any configuration, if a problem arises, the cluster software will be able to perform a local recovery first. Local recovery is the ability to automatically restart applications or services at the local node in the event of a failure. In the case of a node that is not fatal, logical local recovery is the preferred method because it interrupts fewer users than switching to another node.

On the type of failover, some cluster products can be restored in parallel, in which resources can failover to remote nodes in different regions. This is well suited to disaster-recovery requirements. Time, in order to solve the problem of multiple node failures, some cluster products can be cascaded failover, the way it works like a Domino: node one failover to node two, node two failed to switch to node three and so on.