【Global Cloud Observation | Technology Hot Spots】
What are the biggest challenges facing the field of computing power in the current AIGC era?
For this question, the benevolent will see the benevolent, and the wise will see the wise. But the answer that attracts the most attention of the industry is computing power scheduling. Why?
"Born from time": the explosion of multiple heterogeneous computing power has given birth to a computing power scheduling artifact
The emerging technology of AI large model is accelerating the new reshuffle of various industries, of course, there must be a process of industry reshuffle, and in this unprecedented innovation and change, computing power has become the core support. However, the development of computing power faces many challenges.
On the merits, it is better to give an example to be more convincing.
Taking the construction of a smart city computing power overall scheduling platform as an example, there is a large gap between the supply and demand of computing power, and the distribution of existing computing resources is uneven, which not only distributes resources, but also makes it difficult for a single computing power supply model to meet the personalized needs of computing power in various application scenarios. At present, in the critical period of the development of the digital economy, thousands of industries are striving to achieve digital and intelligent transformation, and the application of large model industries is also accelerating.
At the same time, the overall promotion of the construction of urban computing power Internet can realize the efficient matching of urban computing power demand with the efficient supply and demand of computing power resources of national hub nodes, and to a large extent, can also effectively reduce the cost of computing power.
In addition, with the increasing requirements for the supervision of computing power operation, in order to improve the level of computing power operation and supervision, and strengthen the dynamic supervision and evaluation of the city's computing infrastructure, it is necessary to have a unified computing power platform to achieve it.
Based on these challenges and needs, the city has built its own smart city computing power scheduling platform, further strengthened the management of the whole life cycle of computing infrastructure design, construction, and operation, anchored the positioning of "one-network computing power, overall planning integration, and one-stop scheduling", and realized the initial computing power ideal of diversified supply, strong computing empowerment, ubiquitous connection, and safe integration.
Global cloud observation and analysis found that the vision of the city's smart city computing power overall scheduling platform construction is very clear, which is to give full play to the role of the computing power overall scheduling platform, connect the supply and demand of computing power, optimize resource allocation, promote market fairness, ensure service quality, safeguard user rights and interests, promote the construction of industry standards, and create a safe, credible and inclusive computing service innovation platform. The effect of the current promotion has begun to appear, and it is believed that with the acceleration of the development of the industry, more supply and demand sides will benefit. Based on the platform's empowerment of thousands of industries, the city will further optimize the layout of computing infrastructure in the future, strengthen the supply and scheduling of diversified computing power, and improve the level of computing power empowerment services.
With practical actions, we will lead the development of computing power and cloud computing, and successfully build a smart city computing power overall scheduling platform. Exploring the power behind its success is inseparable from the innovative assistance of China Telecom e Cloud, and it is inseparable from the "Xiyang" computing power scheduling artifact.
In fact, not only does the city have a strong demand for a computing power scheduling platform, but in the current AI era, thousands of industries have a direct demand for multiple heterogeneous computing power coordination and scheduling brought about by AI remodeling.
First, with the new digital infrastructure in full swing, general computing power, intelligent computing power, high-performance super computing power, and edge multiple computing power have also prospered, which prompts industry users to face diversified computing power service providers in the choice of computing power, that is, from telecom operators, cloud vendors, data center operators, computing equipment suppliers, etc. to provide diversified computing resources.
Second, while domestic GPUs and CPUs are booming, industry users are facing different computing architectures at home and abroad in specific application scenarios, and how to realize efficient scheduling between computing power of different architectures under the diversification of computing power architectures to achieve real interconnection.
Third, the national strategic project of "Eastern Data and Western Computing" is underway, and there are certain challenges in realizing cross-regional and long-distance computing power grid connection and scheduling, and how to ensure bandwidth, latency, and high reliability.
In terms of solving the cross-service provider, cross-architecture, and cross-regional computing power scheduling barriers, e Cloud's self-developed computing power distribution network platform "Xiyang" has done it. In response to the needs of the AI era, e Cloud has created a computing power scheduling artifact, and "Xiyang" has truly realized the integration of multiple computing power.
"Follow the trend": To build computing power distribution and scheduling capabilities, what is the difference between "Xiyang"?
In order to ensure the smooth distribution and scheduling of computing power, from the very beginning of research and development, e Cloud has followed the standard requirements of ultra-wide coverage, ultra-high reliability, ultra-low latency, ultra-large speed, and cloud-network integration, in order to meet the extreme demand for computing power in thousands of industries.
Under the policy guidance of the "Implementation Opinions on Accelerating the Construction of a National Integrated Computing Network for the In-depth Implementation of the "Eastern Data and Western Computing" Project, it is logical to build a computing service platform that integrates general computing, intelligent computing, and supercomputing resources, strengthen the interconnection and unified services of multiple computing power, and promote the intelligent supply, scheduling, use and settlement of computing power.
Taking the policy as the guide and the industry trend of the computing power Internet as the direction, e Cloud has created a computing power scheduling artifact, and "Xiyang" can be said to be born at the right time and take advantage of the trend. However, in recent years, many ICT vendors in the industry have also launched a variety of computing power scheduling platforms. So, what is so different about "Siyang"?
Computing power access, innovation and breakthroughs. A series of innovative breakthroughs have brought a new atmosphere to the industry's computing power scheduling.
The R&D experts of e Cloud Technology Co., Ltd. once introduced that the first "roadblock" of computing power scheduling is standardization. There are various computing architectures at home and abroad, such as GPU, CPU, DPU, etc., each with its own rules and reasons, and the primary task to realize the future of the computing power Internet is to realize the standardization of heterogeneous computing power access.
In this regard, e Cloud R&D computing power plug-in realizes the standardization of heterogeneous computing power access, and the R&D computing power gateway supports the cloudification management and secure access of idle computing power in the society. At the same time, a computing power measurement system is also established, so that heterogeneous computing power can be measured in a unified manner. Based on this, after receiving the computing power demand, Xiyang can orchestrate and schedule resources according to the real-time perception of the computing network status to provide the optimal solution for computing power. This not only brings direct help to industry users with diversified heterogeneous computing power, but also provides the most powerful platform and tools for computing power optimization.
Computing power service, three-pronged. In order to improve the ability of computing power services, e Cloud has developed computing collaboration technology, after all, computing power ultimately serves data, which helps computing power to better support the insight of data value. At the same time, AI guidance is implemented to help industry users achieve more optimal computing power selection and supply. Moreover, it is worth mentioning that e Cloud breaks through the multi-level computing power scheduling technology to achieve more flexible and intelligent network support, and strengthens the perception of scheduling for the construction of multi-level computing power network between nodes, nodes, clusters, and regional cities.
Computing power ecology, interconnection. Since it is to build a computing power scheduling artifact, it is naturally indispensable to integrate the capabilities and resources of the computing power demand, supply, and operator side, and only by breaking the previous barriers can we build an open computing partner ecology in a real sense and realize the convergence, flow, and sharing of computing resources. In terms of "accelerating the formation of a national integrated computing power system and cultivating the computing power industry ecology", "Xiyang" introduces high-quality third-party computing power through the computing partner win-win plan, provides one-stop intelligent and super-edge computing power services, and is committed to building a national computing power network.
It is worth noting that in addition to providing a public e Cloud self-operated computing power service platform, "Xiyang" also provides a privatized regional computing power interconnection platform for various places, which has been commercialized in many cities.
As a result, it gathers and perceives multiple computing power in real time, including general computing, intelligent computing, supercomputing, and edge, and conducts unified management and scheduling of computing power across service providers, regions, and architectures, so as to achieve efficient matching of multiple heterogeneous computing power supply and demand.
After the upgrade in May 2024, "Xiyang" has overcome a series of key technologies such as computing power plug-ins, computing power gateways, and arithmetic collaboration, and supports the connection of third-party computing power with e Cloud's own computing power, so as to achieve more inclusive intelligent computing. Further analysis shows that in terms of six main indicators and performance, the upgraded computing power distribution network platform "Xiyang" is far ahead of the industry.
In terms of computing power scheduling capabilities, it has supported unified access and scheduling of computing power from mainstream cloud vendors, supercomputing vendors, and intelligent computing vendors in the industry.
In terms of computing power transactions, the operation service layer is a scenario-oriented computing power transaction entrance, and the computing, storage, network, and other resources for high-quality business service requirements have been integrated in supply.
In terms of scheduling algorithm and accuracy, based on the accuracy and efficiency of the algorithm, the algorithm supports self-optimization to meet the scheduling needs of different scenarios.
In terms of timely response rate, millisecond-level latency has been achieved, so as to achieve lower latency in computing power scheduling for industry users.
In terms of timeliness, millisecond-level network-wide policies take effect, which means that "Xiyang" achieves better real-time performance in scheduling computing power.
In terms of platform security and reliability, it provides cloud-native high-availability architecture, data transmission and storage encryption, and provides trusted, traceable, and recordable transaction management.
Although the main indicators and performance lead the industry and are worth paying attention to, the biggest difference of "Xiyang" is its lofty future ideal, so that the computing power of the whole network can be obtained as desired.
"Riding on the momentum": Realize the integration of computing power and accelerate the high-quality development of the computing power industry
Once upon a time, there has always been an ideal in the industry, so that computing power can be used as basic energy sources such as water and electricity, and can be requested and supplied according to the amount to achieve flexible and efficient elasticity.
With its independent R&D innovation ability and excellent computing power scheduling ability, "Xiyang" has been widely recognized by the industry and successfully selected as one of the "Top Ten Super Projects of Central Enterprises in 2022" by the State-owned Assets Supervision and Administration Commission of the State Council.
Obviously, e Cloud has consistently adhered to scientific and technological innovation.
In this regard, Yan Zhiyong, deputy general manager of the intelligent edge business department of e Cloud Technology Co., Ltd. and general manager of Xiyang product line, said that the interconnection of computing power and the integrated scheduling of computing network are the key means to solve the efficient flow of data, optimize the layout of computing resources and data flow, which is also the inevitable development trend of the construction of digital China.
Facing the future and taking advantage of the momentum. Global Cloud Observation and Analysis believes that the "Xiyang" computing power distribution network platform built by China Telecom e Cloud has brought important reference and industry value to the development of the computing industry.
First, global computing power and interconnection. The biggest barrier to computing power comes from service providers, technical architecture, and cross-regional, and through innovative technologies and services, the company realizes the interconnection and scheduling of global computing power, and matches the optimal computing network resources for applications through the cross-domain scheduling capabilities of computing network awareness and on-demand autonomy. AT PRESENT, THE SCHEDULING PERFORMANCE OF A SINGLE CLUSTER EXCEEDS 2000+ INSTANCES PER SECOND, AND THE SCHEDULING SCALE OF THE ENTIRE NETWORK REACHES 20EFLOPS.
Second, realize the integration of computing power and accelerate the high-quality development of the industry. Through the "Xiyang" platform, e Cloud is committed to building an infrastructure for the integrated scheduling of computing power and the efficient allocation of computing resources across regions, empowering thousands of industries to deepen the digital and intelligent transformation, continuously releasing the kinetic energy of diversified computing power, and promoting the high-quality development of the computing power industry.
The wide range of application fields includes the general computing resource pool scheduling that is of most concern to the industry, the Eastern Data and Western Computing Scenarios of Eastern Data and Western Computing, Eastern Data and Western Storage, and Eastern Vision and Western Computing Scenarios, as well as cloud rendering services, AI distributed training, GPU virtualization scheduling, and GPU resource pool scheduling.
Third, Guoyun empowerment and strategic support. The "Xiyang" platform fully supports and serves the national strategic project of "Eastern Data and Western Computing", and provides inclusive computing services for the society, so as to better contribute to the realization of the dual carbon goals.
It can be seen that the future road will be wider and wider, and the future road will be wider and wider.
Moreover, the curtain of multi-heterogeneous computing power scheduling has just begun.
- END-
you
how
Yes
see
？
Welcome to add comments at the end of the article!
【Global Cloud Observation|Technology Talk|Global Storage Observation|Amin Observation】Focus on the analysis of technology companies, speak with data, and take you to understand technology. This article and the author's reply represent personal views only and do not constitute any investment advice.