The concept of a server cluster misunderstanding

What is Cluster? There are two common Cluster (clusters) architectures, one is Web/internet Cluster system, which is to place data on different hosts, that is, multiple hosts are simultaneously responsible for one service, and the other is the so-called parallel operation! Parallel operation is actually the work of the same operation, to the entire Cluster inside the CPU to perform a synchronous operation of a function. With the ability to use multiple CPUs, you can speed up the operation. At present, more common in parallel operations, it is often necessary to see on the supercomputer that these supercomputers are used primarily in the work of astronomical, military, physical, and so much-needed computations, and that, given stability, they are usually used on the hardware architecture above the Unix system.

Like most ideas that have been around for a long time, the concept of server clusters is also hidden in the mist. But the mist will eventually disperse and the concept will come to the bottom.

On the surface, the concept of clustering is not difficult to understand. Clustering is a new technology that joins multiple computers together to form a whole, thus making better use of device resources. Ira Kramer, marketing director of Infinicon Bae, said: "A server cluster joins multiple physical servers in a variety of cluster architectures to work in a distributed cluster." The company provides cluster management equipment and services.

The mystery remains understandable. When technical terminology is passed between engineers and marketers, terminology often loses its original accuracy. At present, there are many similar and related concepts, such as Fibre Channel (fabric), mesh (grid), diffusion (pervasive) and mesh computing (mesh computing), and so on, these methods can improve computational efficiency. Although they mean different things, these words can often be used interchangeably.

A cluster consists of several servers interconnected by an input/output system. These servers are connected to storage media and are managed by distributed resource Management software (DRM). Its components are constantly being improved: Blade servers, fast InfiniBand I/O technology and more complex DRM software, which are grouped together to make clustering a utility for IT managers.

Vision FX, a highly available product program Manager Reier Torgerson, said: "Clustering will undoubtedly become the mainstream technology, the system must be on the power of the time is increasing, we know: Planned and unplanned downtime must be increasingly low." "

Different definitions

There are differences in how different companies name and use these terms, so the exact definition does not exist. Sun focuses on grid and cluster salutation, and Peter Ffoulkes, the planning manager of the company's High-performance Computing group, defines the cluster as a localized computer or server group. The grid is the matrix in which the cluster resides. Grids can exist across regions. For example, Ffoulkes said: The Sun's grid has leapt across California, Texas and Massachusetts. If needed, the company's DRM software can send a task from one cluster to another, rather than running the task locally.

HP focuses on the independence of the grid concept. "The grid is a collection of loosely connected computers with considerable distribution and independence," said Dan Cox, HP's Linux cluster program manager. It is not exactly configured in a clustered fashion. "

"Even after the current downturn, the new economy is changing its business model, which is increasingly supporting clustering, and it is moving from a cost center to a service center," Marty Ward, marketing director of Veritas Software, said.

In the service center model, the IT department manages the resources used by other departments to be more flexible and efficient in managing the methods. This requires technology that can expand its existing resources. Infinicon's CEO, Chuck Foley, said: ' A cluster can increase server efficiency by 15% to 80%, according to a Sun study. "To move to a service model that automates the management of resources and architectures, there are already technologies to achieve that," Ward said. "Although Ffoulkes is unfamiliar with the specific areas of research that Foley refers to, he says:" This data seems reasonable.

Two types of clustering

"In fact, there are two uses for clustering," says Lee Johns, head of global business for HP's industry standards service. One use is to improve the availability of a particular application or service, and the other is to improve the performance of the application. These two usages have different requirements. "

Johns said: "A growing area of demand in clusters is redundant, at a basic level, people do not want the application to fail, as hardware costs decline, security costs become more and more affordable."

Another task of clustering is to provide powerful processing power for specific applications or problems. The goal is to process more information at a given time period or to process it in a shorter amount of time. There are two approaches in the cluster area: "Scaling out" refers to distributing the workload of a given application between servers, and "scaling up" refers to the ability of the runtime to enhance a single server's computing power.

"It is possible to use both methods at the same time," Ffoulkes said. For example, a car manufacturer might want to use clustering to solve the problem of rapidly designing complex new cars. At the same time, the carmaker may also need a database with potential project failure protection mechanisms, so that when important deadlines are approaching, highly paid engineers are not idle. The two cluster operations will be performed separately, Ffoulkes said: "High-availability software is completely different from productivity-enhancing software." "

Technological innovators have a controversy over the economic advancement of clusters. The goal is to provide IT managers with a more flexible ability to cluster. In the past, redundant clusters were performed on a one-to-one pairing between servers. The new software approach has the flexibility to change the ratio of backup home servers.

NSI Software, the company's business continuity director, Jason Buffington, said: This makes the enterprise smart about how to configure the cluster. With this technology, the original one-to-one backup was broken and replaced by a new concept of redundant backups by the enterprise on demand.

Buffington said: "For example, there are 100 servers, we all agree that two to three are critical, need to keep the boot, the other 8 to 10 units may be critical to individual parts." The biggest misconception is that there is a redundant backup of the servers you want to use, which in most cases is not worth it. "

HP's Cox also shares the view that IT managers are now in the decision-making position, saying: "Decisions should be based on the critical level of the application." "

Progress in other areas

Provide additional flexibility to the system and not just the innovations mentioned above. Foley says: Server blades can make hundreds of or even thousands of servers in one room, especially for fibre channel types such as clustering. In addition, the InfiniBand connection to the server at 10Gbps speed, far faster than the previous interface. Cox referred to Myrinet and Gigabit Ethernet and other advanced interconnect interfaces. Finally, 9i RAC and IBM's DB2 EEE databases, such as Oracle companies, are designed for use in Fibre Channel environments. "The biggest difference is that a single application can span multiple servers and remain synchronized at an increasingly large server level," Foley said. "

Server clusters also become more flexible on another occasion. The various operating systems across the enterprise mean that the clusters work together. "What we know is that there are more cross-platform or multiple operating system environments," Torgerson said. We believe that the next big trend is how to collaborate with clusters. For example, how to run an integrated or distributed application when an OS 400 cluster coexists with a cluster in a Microsoft or Linux/unix environment. "

"The clusters of different operating systems will be separated, but will be linked through shared storage media," Cox said.

Developments in other areas are still ongoing. Ffoulkes said: The grid will eventually be internationalized. In this way, a request in New York to calculate the operation may be completed in Beijing. Veritas's Ward said: Veritas is working on automatic discovery Technology, which automatically identifies and configures servers in a cluster properly.

Foley is optimistic about the future of the cluster. "Technically, we're more willing to work with a particular server, but it's too restrictive to bind the application to a particular server or bind the database to a specified server," he said. "