Load balancing Technology: using 4 Web servers

&http://www.aliyun.com/zixun/aggregation/37954.html ">nbsp; The concept of clustering is easily confused with concepts (SMP, NUMA, MPP, distributed processing), whose primary difference is in the level at which resources are shared and replicated. They are arranged by SMP, NUMA, MPP, clustering, and distributed processing from the most compact to the most loosely aligned.

With the increase of traffic volume and the rapid growth of traffic volume and data flow, the processing capacity and computing strength of the existing network have been increased correspondingly, so that single server equipment cannot afford it at all. In this case, if you throw away an existing device to do a lot of hardware upgrades, this will result in the waste of existing resources and, if faced with the next increase in business volume, this will lead to a high cost of hardware upgrades again, and even the most performance-enhancing equipment will not meet the current business growth requirements.

A cheap, effective and transparent way to extend the bandwidth of existing network devices and servers, to increase throughput, enhance network data processing capabilities, and improve network performance, configure the same name for multiple addresses in DNS, so the client that queries this name gets one of the addresses. So that different customers access to different servers, to achieve load balancing purposes. DNS load balancing is a simple and efficient method, but it does not differentiate between servers and the server's current state of operation.

SMP (multi-processor system): This system is in a computer with multiple CPU,CPU status is equal, they share the memory space and I/O devices. It works by operating system to break a task into multiple concurrent processes and then to run it on a different CPU.

NUMA (non-Uniform memory access): A system that allows CPUs to share local memory more efficiently than SMP, which allows faster access to a single area of memory, but can also indirectly access memory in other areas if needed. This approach allows some CPUs to have greater priority in the given range of physical memory.

MPP (Mega parallel Processing): The nodes of this system have their own CPUs and have their own proprietary resources. This structure is relatively independent, but the nodes generally do not have the ability to fully access I/O.

Cluster: Cluster system is composed of independent computers, but the control management tools are unified management.

Distribution processing: It is more loosely connected than the cluster system we are building, typically the task is done in different places, and there is no single entity that can be managed as a whole.

The above polymerization methods are tight and sparse, they have their own scope of application, there is no more to say, interested in looking for some information to see, here just want to let everyone know where it is located.

Ways to achieve load balancing

The purpose of clustering is to share and efficiently utilize resources, provide large operations, provide load-balanced allocation request pressure, and enable switching for high availability in the event of a failure.

Limited to space, this article only introduces the implementation of load balancing (for Turbolinux Cluster Server). Through the analysis of related software, the function of cluster load is realized through traffic management, and there are several implementation methods: direct routing (directly forwarding), network address translation (NAT), tunneling Technology (tunneling).

Direct forwarding

This method can be used when the computer participating in the cluster is in the same network segment as the computer controlling the control, and the computer that controls the management receives the request package and sends it directly to the node participating in the cluster. The advantage is that the flow back to the customer is not controlled by the host, the faster the cost of less.

Network address translation (NAT)

This method may be familiar to everyone, the address converter has a legitimate IP address that can be accessed by the outside world, and it modifies the address of the outgoing packet from the proprietary network, which appears to be from the address converter itself, and when the outside package is sent to the converter, it can determine which node should send the packet to the intranet. The advantage is that the IP address can be saved to disguise the interior; The disadvantage is inefficient, because the traffic returned to the requester passes through the converter.

Tunneling Technology (tunneling)

This method is the forwarding mechanism that is available when the nodes of the cluster are not in the same segment, is the way to encapsulate IP packets in other network traffic, and for security reasons, VPN in tunneling technology or leased line can be used.

The services that the cluster can provide are TCP/ip-based Web services, mail services, news services, DNS services, proxy servers, and so on, and we will implement a load-balanced clustering system for specific products Turbolinux Cluster server. Used to provide Web and FTP services.