How enterprises should choose their own blade server cluster

Computing the number of cluster projects in Linux is like counting the number of startups in Silicon Valley. Unlike Windows NT, which is already hampered by its own closed environment, Linux has a large number of cluster systems to choose from, suitable for different purposes and needs. But it is not easy to determine which cluster should be used.




Part of the reason for the
problem is that terminology clusters are used in different contexts. IT managers may be concerned with how to make the server run longer or to make the application run faster, while mathematicians may be more concerned with large-scale numerical computations on the server. Both require clustering, but they require clusters of different characteristics.





This article investigates different forms of clustering and a subset of many implementations that can be bought and available in free software. Although not all of the solutions listed are open source, most software follows the common practice of distributing Linux source code, especially since those who implement the cluster often want to adjust system performance to meet the needs.





Hardware





clusters always involve hardware connections between machines. In most cases today, this simply means "Fast Ethernet" network adapters and hubs. But in the cutting-edge science field, there are many network interface cards designed for the cluster.





They include myricom Myrinet, Giganet clan and http://www.aliyun.com/zixun/aggregation/14477.html ">IEEE 1596 Standard Scalable Uniform interface ( SCI). The capabilities of those cards not only provide high bandwidth between nodes in the cluster, but also reduce latency (the time it takes to send messages). Those delays are critical for exchanging state information between nodes to keep their operations synchronized.





myricom





Myricom provides network adapters and switches with a maximum one-way interconnect speed of up to 1.28 Gbps. There are two types of network cards, copper and fiber type. A copper-type LAN can communicate at full speed at a distance of 10 feet while operating at half speed for up to 60 feet. The fiber-optic myrinet can run at full speed on a 6.25-mile-long single-mode fiber or a 340-foot-long multimode fiber. Myrinet provides only direct point-to-point, hub-or switch-based network configurations, but there is no limit to the number of switched fibers that can be connected together. Adding switched fibers only increases the latency between nodes. The average latency between two directly connected nodes is 5 to 18 microseconds, much faster than Ethernet.





Cluster Type




The three most common cluster types for
include high-performance scientific clusters, load-balanced clusters, and high-availability clusters.





Science Cluster





Typically, the first involves developing parallel programming applications for clustering to solve complex scientific problems. This is the basis for parallel computing, although it does not use a dedicated parallel supercomputer, which consists of 10 to tens of thousands of independent processors. But it uses commercial systems, such as a set of single processors or dual-processor PCs linked through high-speed connections, and communicates on the public messaging layer to run parallel applications. As a result, you often hear about the advent of a cheap Linux supercomputer. But it is actually a cluster of computers that has the same processing power as a real supercomputer, and typically has a more expensive set of cluster configurations than $100,000. This seems too expensive for the average person, but it's cheaper than a dedicated supercomputer worth millions of dollars.





Load Balancing cluster





Load Balancing clusters provide a more practical system for enterprise requirements. As the name implies, the system allows the load to be distributed as evenly as possible in the cluster of computers. The load may be a balanced application processing load or network traffic load. Such a system is ideal for a large number of users running the same set of applications. Each node can handle a portion of the load and can dynamically allocate the load between nodes to achieve balance. The same is true for network traffic. Typically, a Web server application accepts too much network traffic to be processed quickly, which requires that traffic be sent to a network server application running on another node. It can also be optimized based on the different resources available on each node or on the special environment of the network.





high-availability Cluster




The
high-availability cluster appears to make the overall service of the cluster as available as possible to consider the error of computing hardware and software. If the primary node in the High-availability cluster fails, the secondary node replaces it during that time. The secondary node is usually a mirror image of the primary node, so when it replaces the master node, it can take over its identity completely and thus make the system environment consistent with the user.





is often mixed and mixed between these three basic types of clustering. As a result, high availability clusters can also be found to balance the user load among their nodes while still trying to maintain high availability levels. Similarly, you can find a parallel cluster from the cluster that you want to program into, which performs load balancing between nodes. Although the cluster system itself is independent of the software or hardware it is using, the hardware connection will play a key role in running the system efficiently.





Giganet





Giganet is the first supplier of the Virtual Interface (VI) architecture card for Linux platforms, providing clan cards and switches. VI Architecture is a platform-independent software and hardware system developed by Intel to create clusters. It uses its own network communication protocol to exchange data directly between servers, rather than using IP, and it is not intended to be a WAN routable system. Now, the future of VI depends on the ongoing work of the system I/O Group, which is a merger of Intel-led "Next generation I/O" teams with "future I/O teams" led by IBM and Compaq. Giganet products currently provide 1 Gbps one-way communication between nodes, with a minimum delay of 7 microseconds.