Clustering-Fundamentals 1

Background

With the rapid increase of Internet access, the ability of a single server has been severely unable to meet the demand. You need to consider improving service capability from two aspects: 1, scale up, 2, scale out

Disadvantages of scaling up:

1. High cost

2, with the improvement of performance, will encounter bottlenecks at a certain point, resulting in lower performance.

The benefits of scaling out:

1. Low cost

2. Provide high concurrency capability and high availability

3, scalability is good.

Classification

Load Balancer cluster (load Balance)

Highly available clusters (high availability Cluster)

High performance cluster (performance computing)

Load Balancing cluster: Due to the reason of the telco's DNS, it is not possible to do load balancing by DNS, and it must be distributed to the backend servers through the front-end scheduler to increase concurrent traffic. However, with the increase of traffic, scheduler and storage, etc., access speed will become a bottleneck, so the site needs to be functional segmentation, cluster, such as the classification of the portal site. To ensure the consistency of content provided by the backend server, the static content of the Web page is synchronized using the rsync+inotify mechanism. A Web page contains multiple Web objects, and the current browser supports multiple threads, so requests from clients, possibly multiple requests, and dispatches to multiple back-end servers, can increase the speed at which the Web page is opened. And the dispatcher has the ability of health check, can remove the host that is hung, also can add the restored host to the cluster.

Advantages: Improve concurrency processing power

Highly available clusters: can improve the online capabilities of services, service availability. For example, there are two hosts to provide Web services, a as a normal service host, B as an alternate host, a constantly send their heartbeat information by multicast to B, and B can also check a heartbeat information. When b cannot get the heartbeat of a, it turns off A's power through the power switch and takes the IP address of a to the service provided before a to improve the online capability of the service. If more than one host's high-availability cluster, you can also set the priority of the standby node, in response to the main node after the action of hanging off.

High-availability clusters deliver more than just heartbeat information, as well as transactional information (priority, etc.). The node responsible for coordinating the completion of the transaction is called a DC (one of the nodes), and if the DC is hung up, one of the other hosts can be chosen.

Key differences between a load-balanced cluster and a highly available cluster:

Load Balancing cluster provides concurrent processing power, providing health check

The always-on-line capability of a highly available cluster to provide services heartbeat

High-performance cluster: The Distributed File system is used to decompose complex problems into small problem parallel processing.

Because the efficiency of a highly available cluster is not very high (a backup host can cause a waste of resources), it needs to be improved. For example, Web services and mail services are started on A and B hosts, and then they are primarily prepared for each other, improving efficiency. However, a highly available cluster can only be started on a single host at a time.

brain fissure: because of cluster division, when a node in a cluster cluster temporarily stops responding because of a processor busy or other reason, the other node may mistakenly consider the node "dead", thereby seizing access to the shared disk (that is, the resource). At this point it is very likely that the suspended animation node re-writes to the shared file system, causing the shared disk file system to become corrupted.

Stonith: Headshot, shoot the other node in the head switches off the power of the host that detects no heartbeat information through the power switch.

Quarantine: fancing, denying a node access to a resource. It is divided into node isolation (Stonith) and resource isolation.

To prevent the cluster from splitting, the cluster node should be 3 or more than 3 odd numbers.

Other knowledge:

DAS: Direct Attach storage. The kernel directly operates data for block devices. A write error can occur if different hosts access the same file at the same time. But the efficiency is high.

NAS: Network Attached storage. is a file-level operation, when the first host writes data, a lock is applied, and the second host cannot write. Performance is worse than Das.

This article is from the "Walker" blog, please be sure to keep this source http://223228686.blog.51cto.com/2222284/1716879

Clustering-Fundamentals 1