Preface to the present, both in the corporate network, intranet or in a wide area network such as the Internet, the development of the business volume exceeded the most optimistic estimates of the past, the Internet has rolled out an endless stream of new applications, even if in accordance with the optimal configuration of the network, you will soon feel too much.
Especially in the central part of the network, their data flows and calculation of strength, makes a single device cannot assume, and how to do the same function for multiple network devices to achieve reasonable traffic distribution so that it will not appear one device is too busy, and the other device do not fully exploit the processing power, becomes a problem, the load balancing mechanism also came into being. Load Balancing is built on top of the existing network structure, it provides a cheap and effective way to expand Server bandwidth and increase throughput, enhanced network data processing capacity, improve network flexibility and availability. It is mainly to: resolve network congestion, provide services, location-independent; provides users with better access to quality; increase server response times; increase server and other resource use efficiency, avoid network key parts of a single-point failure. Define actually, load balancing is not a traditional sense of the word "balance", in General, it just may be congestion to one local load to multiple local share. If you change it to "load-sharing", perhaps better understand some. Very popular, and load balancing across a network effect is like taking turns on duty system, the tasks assigned to you to complete, so as not to allow a person to death the fatigues. However, this sense of equilibrium in General is static, that is, to a predetermined "rotating" policy. And take turns on duty system, dynamic load balancing through some of the tools in real time analysis of packets, mastering network data traffic situation, reasonable distribution of the tasks. The structure is divided into local geographical load balancing and load balancing (global load balancing), the former refers to the local server clusters do load balancing, the latter is in a different geographical location, in a different network and server for load balancing between clusters. A server cluster node for each service you want to run an independent copy of the server program, such as Web, FTP, Telnet or e-mail server program. For some services (such as running in the Web server on the service), a copy of the program running on all hosts within the cluster, network load balancing the workload between these hosts. For other services (such as e-mail), only a single host handling workload, for these services, the network load balancing allows network traffic flow to a host, and in the host fails to traffic moved to a different host. Load balancing technology to implement on top of the existing network structure, load balancing provides a cheap and effective way to expand Server bandwidth and increase throughput, enhanced network data processing capacity, improve network flexibility and availability. It is mainly to complete the following tasks: ◆ resolve network congestion problem, provide services, location-independent ◆ provides users with better access to quality ◆ improve server response speed * increase server and other resource use efficiency ◆ avoid network key parts of a single-point failure broadly load balancing you can set up a dedicated gateway, load balancers, or some special software and protocol implementations. On a network load balancing applications, from a network of different levels, depending on the network bottleneck for specific analysis. From the client application as a starting point for the analysis of longitudinal, refer to OSI's tiered model, we have implemented load balancing technology is divided into client-side load-balancing technologies, application server technologies, high-level protocol switching, network access protocol exchange, and other ways. Load balanced hierarchy ◆ client-based load balancing this mode refers to the network client runs a specific program, the program through the regular or occasional collection server farm running parameters: CPU usage, disk IO, memory, and other dynamic information, and then under a selection strategy, find the best available service, the local server application requests made to it. If you load information collection program found server failure, you find other alternative server as a service. The entire process for applications is completely transparent, all work is processed at run time. So this is a dynamic load balancing technology. But this kind of technology there are interoperability issues. Because every client that you want to install this special collection procedures; and, in order to ensure transparency in running the application tier, the need for each application be modified through the dynamic link library or embedded method, the client's access request to go through acquisition program and then sent to the server to redirect the process. For almost every application to the development of the code, the workload is relatively large. Therefore, this technology is only in special applications, such as is used to perform certain specific tasks, more needs to be distributed computing power, the application developer does not have too many requirements. In addition, the use of JAVA framework model, often use this model of load balancing for distributed, because the application is based on java virtual machines, you can at the application layer and the virtual machine design an intermediate layer between the processing load balanced work. ¡Ô application server load balancing technology if the client load balancing layer to an intermediate platform to form a three-tier architecture, the client application may not need to make special modifications, transparent through middle-tier application server will request equilibrium to the appropriate service nodes. Common implementation means that the reverse proxy technology. Use a reverse proxy server, you can even request forwarded to multiple servers, or directly to the cached data is returned to the client, this acceleration mode to a certain extent, can improve the access speed of static Web pages, to achieve load balancing purposes. The benefits of using a reverse proxy is that you can load balance and the proxy server's cache technologyCombined together, provide a useful performance. However it also has some problems, first is a must for every kind of service is specialized in the development of a reverse proxy server, this is not an easy task. A reverse proxy server itself while it is possible to achieve very high efficiency, but for every proxy, the proxy server must maintain two connections, one external connections, an internal connection, therefore particularly high connection requests, the proxy server load is very high. Reverse proxies can perform optimized for application protocol load balancing policy that each time only access most idle internal servers to provide services. But as the number of concurrent connections increases, the load on the proxy server itself has also become very large, and finally the reverse proxy server itself will become a bottleneck. ◆ The domain name system-based load balancing NCSA's extensible Web was among the first to use dynamic DNS polling technology Web system. In the DNS configuration for multiple addresses with a first name, query the name of the client will receive one of these addresses, allowing different clients to access different servers, to achieve load balancing purposes. In many well-known web sites use this technique: including early yahoo site, 163, etc. Dynamic DNS implementation up a simple polling, no complex configuration and management, General support bind8.2 above class UNIX systems are able to run, so widely used. DNS load balancing is a simple and effective way, but there are many problems. First of all a domain name server was unable to know the services node is valid, if the service node fail-over system, continues to be the domain name resolves to the node that caused the user access to fail. Secondly, because the data refresh time DNS TTL (TimetoLIVE) flag, as soon as you exceed the TTL, other DNS servers as needed and this server interaction, to regain the address data, it is possible to obtain a different IP address. Therefore in order to enable the address can be assigned randomly, you should enable TTL as far as possible short, different parts of the DNS server can update the corresponding address, random access to addresses. However the TTL set too short, will allow DNS traffic, but result in additional network problems. Finally, it cannot tell the difference between the server and does not reflect the current running state of the server. When you use DNS load balancing, you must try to ensure that a different client computer can even be different addresses. For example, user A may just browse several pages, and user b may be a large number of downloads, because the domain name system there is no suitable load policy, simply by rotating the balance, it is easy to user A request to load the site, and the light will B requests to load is very heavy site. Therefore, the dynamic properties of dynamic DNS on polling results are unsatisfactory. ◆ High level agreements content exchange technology in addition to the above several load balanced way, there are in agreement with the internal support load balancing capability of the technology, that is, the URL Exchange or seven layer Exchange, provides a high level of access to the flow of control. Web content switching check all of the HTTP header, according to the information in the header to perform a load-balanced decision-making. For example, can this information to determine how to personal home pages, and image data, and other content providers services, common HTTP protocol redirection capability, etc. HTTP runs on top of the TCP connection. Client through constant port number 80 TCP service directly connected to the server, and then over TCP connections to the server sends an HTTP request. Protocol Exchange based content strategy to control the load, and not according to the TCP port number, so no access to the flow of stranded. Due to the load balancing device to access request to multiple servers, so that it can only establish a TCP connection, and HTTP requests through to determine how to load-balance. When a Web site's hits per second over a hundred or even thousands of times, TCP connections, HTTP header information for analysis and process delays have become very important to do everything possible to improve the performance of various parts. In the HTTP request and the header has a lot to load balance the useful information. We can learn from these information in the client that the requested URL and Web pages, the use of this information, load-balancing devices can put all boot image request to an image server, or database queries based on the URL of the CGI program is called the content, the request is directed to a dedicated high-performance database server. If the network administrator is familiar with the content exchange technology, he can HTTP header field to use cookie Web content exchange technology to improve specific customer service, if we can find from the HTTP request, some laws can also make full use of its various decisions. In addition to the TCP connection table problems, how to find the appropriate HTTP header information, and to load-balance the decision-making process, is the impact of Web content switching performance of important issues. If the Web server has SSL for image service, dialogue, database transaction services such as special function has been optimized so that the level of flow control to improve network performance. ◆ Network access protocol exchange large network normally consists of a large number of special technical equipment, such as including firewall, router, 3, 4-layer switches, load balancing, caching servers and Web servers, etc. How to apply these technological devices combine organic land, is a direct impact on network performance of critical issues. Now many switches provide the fourth layer switching function, provides a consistent IP address, and maps to multiple internal IP address on each TCP and UDP connection request, according to its port number, in accordance with the established policy dynamic select an internal address, the packets are forwarded to the address of the load equally. Many hardware manufacturers will this technology integrated in their switches, as they make the fourth layer
No comments:
Post a Comment