Low power consumption small volume: design and implementation of new cloud computing server

1. Overview

Cloud computing is a new computing model whose core driving force is that businesses or individuals no longer need to install a large number of applications on their computers, which are accessed through a Web browser to a wide range of on-demand services, called cloud services. It will radically change the user experience.

The traditional server is mainly oriented to scientific computing, and improves the performance of the server by increasing CPU clock frequency, instruction set parallelism and kernel number. INTEL/AMD/IBM, such as the company's server products are too demanding performance, ignoring the product power consumption, volume, cost and other energy efficiency factors, resulting in lower server energy efficiency, but also caused the server large size, high power consumption problems. Cloud computing is mainly business computing, it does not pay attention to the performance of mathematical operations, but the need to solve a large number of logical operations and multi-user simultaneous request problems. Therefore, we must adopt a new idea to design cloud server system structure. This paper analyzes the system structure of the new cloud computing server, and designs and ICCS-PW a new cloud computing server.

2, the new cloud computing server system structure

The new cloud server focuses on High-performance throughput computing capabilities, focusing on the sum of work over time. Therefore, the new cloud computing server in the architecture and traditional server is very different. In the traditional server, all contain processor touch block, enclosure, network module, power supply, fan and other equipment.

The new cloud server architecture is divided into the Cloud processor module, the network processing module, the storage processing module and the system module, etc. by the function block, they provide the hardware virtualization of the network module and the enclosure by introducing the tight coupling high speed interconnection network. The new cloud computing server model is shown in Figure 1.

Compared with the traditional server architecture, the new cloud computing server has greatly improved its utilization rate, but also omitted a lot of duplicate hardware in the system. According to the characteristics of cloud computing services, using a number of cloud processors to complete the system design, and the introduction of Low-power management concept to complete the system's centralized redundancy management.

The new cloud server integrates computing, storage, exchange and management into a standard system, which is compatible with external network devices, storage devices and common architectures, and the operating system mainly supports Linux for user application software porting.

3, the new cloud computing server key Technologies

3.1 Cloud Processor

The new cloud processor designed in this paper focuses on High-performance throughput computing. High-performance throughput computing is the link between high-end single-threaded performance and multi-core multithreading, focusing on the total amount of work over time, rather than looking at the performance index and results of a single kernel or single thread running on a benchmark.

The cloud server generally includes the total calculation of the function units, such as thread, kernel, processor, network, accelerator, etc. Therefore, cloud computing generally has a huge amount of data input or a large number of working sets, resulting in a lot of memory, but there is lack of access to the local space, so that the existing multilevel cache memory architecture is greatly improved.

Because in practice, many servers drive large amounts of data through hundreds of double-sided built-in memory models, leaving the memory inevitably away from the processor, making it possible for the processor to spend much of its time waiting for data or instructions from distant memory. At the same time, the interruption of cloud processing is more, the relationship between CPU calculation and data access must be handled well.

(Responsible editor: The good of the Legacy)