1. Two key features
A cluster is a group working together service entity (understood as a server) to provide more scalability and availability service platform than a single service entity. In the client view, a cluster like a service entity, but in fact cluster consists of a set of services entities. Compared with a single service entity, clustering provides two of the following key features:
Scalability - Performance Cluster service is not limited to a single entity, a new service entity can be dynamically added to the cluster, thereby enhancing the performance of the cluster.
High Availability - Redundant Cluster service entity by the client from easily met out of service warning. In a cluster, the same service can be provided by multiple service entities. If a service entity fails, another service entity will take over the service entity failure. Clusters provide service from a wrong functional entity to another service entity restore enhanced application availability.
2. Two capacity
For scalability and high availability features, two clusters must have the following capabilities:
· Load Balancing - Load balancing can a more balanced distribution of tasks to computing and network resources in a clustered environment.
· Error Recovery - for some reason, the resources to perform a task fails, the other service resources to perform the same task entity then complete the task. This can not work because of an entity's resources to another entity resources to complete the task of continuing transparent process is called error recovery.
Load balancing and error recovery services are required in each entity have the resources exist to perform the same task, but the same task for each resource, the information required to perform a task view (information context) must be the same.
3. The two technologies
Be sure to have the cluster to achieve the following two technologies:
· Cluster address - a cluster of multiple service entities, the cluster client acquisition function of each service entity within the cluster by cluster address to access the cluster. With a single cluster address (also called a single image) is an essential feature of the cluster. Maintain the cluster address are called a load balancer. Join and leave the internal load balancer is responsible for managing each service entity responsible for external cluster address to the physical address of the internal conversion service. Some load balancer to achieve true load balancing algorithms, and some only support conversion tasks. Achieved only task switching load balancer for supported ACTIVE-STANDBY cluster environment, where the cluster is only one service entity work, when the service entity is operating fails, the load balancer to the latter task turned to another service entity.
Internal communications - in order to be able to work together to achieve load balancing and error recovery, clustering among entities must always communicate, such as load balancing heartbeat test information service entities, the service communication between entities task execution context information.
Has the same cluster address so that the client can access the cluster computing services provided, a hidden internal address of each service entity under the cluster address, making computing services to customer requirements can be distributed among the various service entities. Internal communication is the basis of the cluster can operate normally, and it has the ability to make cluster load balancing and error recovery.
4. Cluster Category
Linux cluster is divided into three categories (high-availability clustering, load-balanced cluster, scientific computing clusters), high availability cluster (High Availability Cluster), cluster load balancing (Load Balance Cluster), scientific computing cluster (High Performance Computing Cluster)
Specifically including 4.1
Linux High Availability Availability Cluster: Common two hot standby node, multi-node HA cluster, RAC, shared, share-nothing clusters and the like;
Linux Load Balance cluster load balancing: LVS etc;
Linux High Performance Computing Clusters of high-performance scientific computing: Beowulf class cluster, distributed storage and so on;
Other categories linux cluster: The Openmosix, rendering farm like ..
4.2 Details
4.2.1. High Availability Cluster (High Availability Cluster)
The common node is made of two HA cluster, there are many popular scientific name, such as "hot standby", "Dual Active", "Dual."
High availability clustering solutions is the ability to protect the user's applications continue to provide services to external. (Please note that neither the high availability cluster is used to protect business data protection is the user's business processes outside uninterrupted service, to reduce the impact of failure due to software / hardware / man-made business to a minimum).
4.2.2. Load balancing cluster (Load Balance Cluster)
Load balancing system: all nodes in the cluster are active, they share the work load on the system. General Web server cluster, database cluster and application server clusters are of this type.
Load balancing clusters are generally used for Web servers, database servers corresponding network requests. Such clusters may, upon request, receiving less inspection request, the server is not busy, and the requests to these servers. This point of view, load balancing and fault-tolerant cluster are accessible from other servers to check status, except that a larger number.
4.2.3. HPCC (High Performance Computing Cluster)
HPC (High Perfermance Computing) clusters, referred to as HPC clusters. Such a cluster is committed to providing powerful computing power of a single computer can not provide.
Extended: High Performance Computing Category
High throughput calculated (High-throughput Computing)
There is a class of high-performance computing, it can be divided into several parallel sub-tasks and subtasks is no correlation between each other. Like at home, the search for extraterrestrial (SETI @ HOME - Search for Extraterrestrial Intelligence at Home) is this type of application. This project is the use of idle computing resources on the Internet to search for aliens. SETI project server data model and a set of data sent to participate in SETI compute nodes on the Internet, computing nodes with a given search mode on the given data, and then the search results will be sent to the server. Server is responsible for bringing together the various computing nodes from the returned data into a complete data. Because a common feature of this type of application is to search for certain patterns in the data mass, the so called high throughput calculation such calculations. The so-called Internet computing fall into this category. Flynn's classification according to, high throughput calculation belongs SIMD (Single Instruction / Multiple Data) category.
Distributed Computing (Distributed Computing)
Another type of high throughput and compute exactly the opposite calculation, although they can be divided into several parallel to sub-tasks, but very close contact between sub-tasks, requires a lot of data exchange. Flynn according to the classification of the genus distributed HPC visible MIMD (Multiple Instruction / Multiple Data) is.