Scale out is literally beyond the size range, while scale up is proportional. That is to say, you can increase the computing power by adding a processor and adding an independent server.
Scalability must be considered for the server system ). Unless the business never grows, as the number of users increases, the server will soon reach the performance and concurrency limits. There are usually two solutions to this problem:
It represents the scale-out of distributed computing and the scale-up dominated by hosts or chassis.Scale-out (scale-out): enterprises can add different server applications as needed, rely on multi-server collaborative computing, and use load balancing and fault tolerance to improve the computing capability and reliability. Scale-up: Large backend servers of an enterprise are upgraded to increase computing resources such as processors to meet application performance requirements. There is no obvious difference between these two technologies today. Each provider not only provides UNIX and Windows platforms for distributed computing, but also UNIX and Windows platforms for centralized computing. Even traditional centralized computing mainframes are proving the nature of distributed computing, as evidenced by the use of Linux and Z/VM Virtual performance on IBM zseries servers. However, larger and stronger servers are also more expensive. The cost is usually higher than the deployment of a large number of relatively inexpensive servers to improve performance. In addition, the server performance can be improved to a certain extent (distributed deployment is relatively higher than the performance improvement ). Therefore, we need to use scale-out to achieve scalability, while allowing users to retain the path to improve system capabilities by adding servers. However, there are also many difficulties to solve in implementation: first, to successfully implement the outward scaling architecture, you must solve the complex distributed computing problem (this problem is not required for the scale-up solution ), the solution to this problem often requires complicated technologies and a relatively large amount of money. large sites such as Google, Yahoo, and Amazon.com all develop a large number of related technologies on their own. Second, the scale out scheme also requires a lot of rewriting work on the software originally used, to ensure that the system can run on the Distributed Server (the scale-up solution requires almost no changes to the existing software ). This step is often a nightmare for developers in every company. A poor performance will waste all the work of developers.
Furthermore, the scale out solution is always faced with the problem of data sets, that is, the split data is still relatively concentrated in the server logic system, rather than unlimited random splitting. If a large number of logics are placed at one end of the database server, the database server will cause the system to lose the capability and possibility of scale out. Therefore, to ensure the scale-out capability, you must ensure that the database only processes substantive data submissions and unavoidable data queries, methods should be taken to avoid data queries and non-substantive data submission that can be avoided. There is no optimal method for specific policies and solutions.
PS:
SMP: symmetric Ric Multiprocessing, that is, symmetric multiple processing. A group of processors (multiple CPUs) are collected on a computer ).
They share the memory and bus structure, and the system distributes Processing Task queues evenly across multiple CPUs, greatly improving the system's data processing capabilities.
As the application level improves, it is difficult for a single processor to meet the actual application requirements. As a result, server vendors have adopted Symmetric Multi-processing systems to solve this conflict.
The most common Symmetric Multi-processing system in PC servers uses 2, 4, or 8 processors. Unix servers support a maximum of 64 CPU systems, such as Sun's product e10000.
The most critical technical problem in SMP systems is how to better coordinate and communicate with multiple processors.
