Hotspot in Linux cluster technology-Virtual Technology

In the IT industry, the development of virtual technology is ready, but IT is difficult to succeed in Linux clusters. Donald Becker, one of the founders of Beowulf, says this is a pity because the system provides only one virtual option that doesn't make any sense.
Becker said that the Linux cluster system should pay more attention to the development of virtual technology and high feasibility, and explained the reasons in his interview. At the same time, he gave some suggestions to users who switched to Linux cluster Symmetric Multi-processing.
Besides his role in Beowulf, Becker is also the founder and main scientist of Scyld software. He has just joined the expert group on this site and serves as a special expert to answer users' questions about Linux Clusters and servers.
Are there any Latest Developments in Linux cluster technology?
Donald Becker: at present, the hot spot in cluster technology is virtual technology. Two famous traditional virtual technology systems are VMWare and Xen. VMWare imitates and simulates the hardware of a personal computer. This allows it to run almost all operating systems in its own OS.
However, one drawback of the development of virtual technology is that it often implies and relies on some large upper-layer technologies. As a result, VMWare has no chance to optimize the system by imitating some upper-layer technologies. Xen is a system management program that runs multiple operating systems at the same time. It is not a comprehensive imitation, thus reducing the dependence on upper-layer technologies. However, it still needs to optimize its core technologies.
What is the difference between the virtual technology on a Linux cluster and other methods?
Becker: Both VMWare and Xen imagine that they can run and fully Install Multiple kernels. Running clusters provides the opportunity not limited by virtual technology, and creates an easy environment where only the necessary conditions for running applications are met. When the system server has a task that requires the main network terminal to complete, then the computer terminal only needs to run the application, which can improve work efficiency, this is also formally different from the traditional virtual system.
◆ Some IT industry schools say they use Symmetric Multi-processing (SMP) systems for high-intensity data computing. What challenges will these technologies face when porting Linux clusters?
Becker: the answer to this question depends largely on the application of the application you are running.
The earliest application to enter the cluster is the Execution System of user parameters. In this model, multiple independent applications run on different data processing machines.
If the Symmetric Multi-processing (SMP) system is applied based on its high efficiency for single-process operations, Beowulf clusters will be a good alternative. However, if the Symmetric Multi-processing (SMP) system is working based on a shared memory of multiple processes, the structure of the application may be complicated.
A traditional application of SMP is a conversion database. This is a special case for fine-grained communication and running on SMP, but the same result can be obtained on the cluster. Therefore, careful analysis is required to determine whether it is feasible to port SMP to the cluster.
Most programs are not compatible with these two extremes.
◆ Can a Linux cluster achieve high availability (HA), or can it only maintain its inherent complexity?
Becker: scalability does not exclude high practicability, so Linux clusters can achieve high practicability (HA ). HA is based on stable support for business applications and data. The normal running time of computer programs should be more than 99.999%, that is to say, a maximum of one second of downtime can be performed in a day.
The classic efficient cluster configuration includes two servers, which can understand the status of each other and share a storage subsystem. If any server fails, the other server will take on all the work. Efficient clusters achieve stable reliability by connecting two or more servers through troubleshooting, load balancing, redundant processing, and other methods, this reduces the inevitable or accidental storage loss and shares storage devices.
Due to their limited absolute capacity, large clusters may frequently encounter problems, so we need to design programs to solve these problems. However, this design will not happen automatically, and many tools on the market do not solve this problem. Well-designed subsystems, such as some commercial evaluation systems with good management capabilities, can handle the increased testability. This means to measure a variety of parameters and avoid problems such as scalability, these problems often occur when a terminal occupies a large amount of storage page or system file data and other terminals cannot work.
◆ In Linux cluster systems, how can we achieve high efficiency?
Becker: in the cluster system configuration, multiple servers are managed through central control points in the same system. This represents a significant advantage in the efficiency status, because many servers can access the backup resources, which allows 24x7 feasibility, error maintenance, and central management of distributed applications, processing capabilities for large data resources, dynamic website publishing, and disaster recovery.
It is important to select an appropriate method to ensure data integrity during the downtime and restart period. Unfortunately, not all options can improve reliability and maintain data integrity. Therefore, it is important to choose your execution program properly at work.
Some non-commercial cluster systems create a local single system concept, which can be completed in the following way: requesting virtual memory in the network; or a global file system; or carrying out a clear system transplant. However, this design cannot properly handle errors because the system must spend a certain amount of time fixing or stopping all programs associated with this terminal.
A good solution is to ensure that the main terminal of the cluster continues to run. For a large installation system, it is necessary to ensure that a major terminal is backed up by conventional equipment and disaster recovery; or multiple major terminals at the same time. In other words, a computer terminal can join or leave a cluster without affecting the basic system, even if applications on it have been stopped.

Related Articles]

• Red Hat Linux new system integrated virtual technology for easy operation

• Virtual Technology will determine the future of the economy

• Virtual Technology helps reduce the pressure on server Growth