Views: 81 Author: Site Editor Publish Time: 2022-02-01 Origin: Site
At present, in order to promote the intensive construction and operation of IT support systems and further play the advantages of centralized capacity, IT cloud has become the infrastructure for the construction of IT support systems for operators. However, in the process of IT cloud resource pool deployment, server technology faces several new challenges, which are mainly reflected in the following three aspects.
In terms of performance, the rapid expansion of artificial intelligence (AI) applications requires the IT cloud to adopt high-performance GPU servers. ai has been applied at scale in scenarios such as network coverage optimization, bulk complaint delimitation, anomaly detection/diagnosis, service identification, and user location in the telecom industry. the emergence of a large number of AI application requirements requires data centers to deploy servers with better computational performance, throughput capacity, and latency performance, with The traditional general-purpose x86 server-based computing platform appears to be overwhelmed, and GPU servers are therefore on the historical stage of operator IT construction.
In terms of efficiency and cost, the IT cloud deployment of general-purpose servers have disadvantages, giving rise to the demand for customized whole cabinet server applications. In the process of IT cloud construction, due to the rapid growth of business demand, IT cloud resource pool expansion pressure, the number of servers in the cloud resource pool is rapidly increasing, the need to improve online efficiency. At the same time, the disadvantages of traditional general-purpose server deployment mode with long cycle time and low deployment density bring greater challenges to data center space, power, construction cost and efficient maintenance management. Whole cabinet servers have become another optional solution for IT cloud construction.
In terms of energy saving, the rapid development of high-density application scenarios such as AI drives liquid-cooled servers to become a hot spot. With the development of AI high-density business applications, the future data center server power will develop from 3kW~5kW to 20kW or even more than 100kW scale, the traditional air-cooled server cooling system solutions can no longer meet the cooling needs, liquid-cooled servers become an effective solution for AI application scenarios.
The development trend of whole cabinet server and its application in telecom industry
The whole cabinet server is a server solution built according to the modular design idea, and the system architecture consists of six subsystems: cabinet, network, power supply, server node, centralized heat dissipation and centralized management, which is a fundamental change to the data center server design technology. The whole cabinet server pools power supply units and cooling units to increase the deployment density by saving space, and its deployment density can usually be doubled. The design of centralized power supply and heat dissipation makes the whole cabinet server only need to configure 10% of the number of power supplies of traditional cabinet servers to meet the power supply needs, and the power efficiency can be increased by more than 10%, and the energy consumption of a single server can be reduced by 5%.
Domestic operators in the IT cloud construction has promoted the deployment of the whole cabinet server, after the actual application test, in the following aspects of the advantages are obvious.
First, factory prefabricated, delivery time is significantly shortened. Traditional server delivery efficiency is low, the use of the whole cabinet server will be the original data center site for the unpacking of servers, shelves, cabling and other work transferred to the factory to complete, the deployment of the granularity from 1 to dozens of units, delivery efficiency greatly improved. Taking the one-time delivery of 1,500 servers as an example, the delivery workload can be reduced by 170~210 man-days, and based on the daily allocation of 10 people, the on-site delivery time can be saved by about 17~21 days.
Second, the pooling of resources brings about a reduction in the number of components and a significant decrease in failure rate. By pooling the resources of power supply, cooling and other components, the whole cabinet server significantly reduces the number of components, bringing a significant reduction in failure rate. Figure 1 compares the number of power supply components and the monthly failure rate of each 32-node whole cabinet server with traditional 1U and 2U server models within one year. Since the 32-node full cabinet server contains 10 power supply components compared to 64 power supply components for 32 1U general-purpose servers, the number of full cabinet power supply components is reduced by 84.4% in comparison. The monthly failure rate of the 32-node full cabinet server relative to the 32 1U general-purpose servers is also significantly reduced due to the lower number of power supply components.
Third, the O&M efficiency is improved by more than 60%. The whole cabinet servers are prefabricated in the factory, network cables are pre-processed in the factory, cable lengths are precisely matched, cable management is simple, wiring is uniform and standardized, and with the operation and maintenance label, it is more convenient and simple to implement maintenance operations on nodes in operation and maintenance, which effectively reduces operation and maintenance mishaps, improves operation and maintenance efficiency by more than 60%, and significantly reduces the fault recovery time after failure.