Cloud computing is born to simplify the use of resources for users, so that users do not have to pay attention to the complex underlying hardware architecture, but call resources through "services" to focus on their own business innovation.
Therefore, if users want to go to the cloud, it is natural that the cloud platform must be able to shield the underlying hardware architecture, and complete adaptive scheduling and unified management of resources according to the characteristics of upper-layer applications. Cloud computing vendors must do their best to do this well.
One cloud with multiple cores , a global proposition
Chips, the foundation of IT infrastructure, provide computing support for the release of IT capabilities.
With the development of technology platforms and business applications such as big data, artificial intelligence, edge computing, and the Internet of Things, different chip architectures are good at different business scenarios, which brings the coexistence of x86, ARM, RISC-V and other chip architectures, and the emergence of SoC, CPU, AIPU and other types of chips. In addition, some customers choose not to put eggs in one basket out of business continuity considerations, and introduce more chip combinations. All of these make the IT environment more and more complicated.
High-tech companies represented by Amazon, Google, and Meta have developed their own chips many years ago. In recent years, it is necessary to carry out the test and application of one cloud and multiple cores.
This is the case with giant companies. How to make users "zero" aware of the complicated chips has become a challenge for all cloud computing providers. One cloud with multiple cores, a global proposition.
The first "one cloud with multiple cores" SPEC Cloud benchmark test, Inspur Yunhai is a global proposition
Compared with foreign countries, the domestic chip environment is more complicated due to demand-driven.
The "many" of "chips" has such a set of statistics: as of May 2021, the number of companies in the domestic AI chip industry alone has exceeded 3,000. More than two years have passed, and as the battle between "100 models" and "1000 models" has become more and more serious, this number will only become more and more terrifying.
The more complicated the environment, the more it can hone a person's will and stimulate his potential. The same is true for businesses. We have seen that public cloud vendors represented by Alibaba Cloud deployed one cloud with multiple cores many years ago and achieved certain results. Recently, facing more complex private cloud scenarios, Inspur Yunhai's "one cloud, multiple cores" has sprung up, completed the industry's first "one cloud, multiple cores" SPEC Cloud benchmark test, established an industry reference benchmark, and promoted the implementation of "one cloud, multiple cores" in multiple industries.
In this regard, Zhang Dong, chairman of Inspur Data, said: "For Inspur Yunhai, 'one cloud with multiple cores' is no longer a simple slogan, but a direction to practice hard."
Recently, Inspur Yunhai OS has completed the world's only SPEC Cloud benchmark test for multi-core scenarios in one cloud. In the mixed deployment cluster test of three processor nodes, the relative scalability and average instance configuration time and other indicators have reached the world's leading level, which verifies the high efficiency, high performance and high scalability of Yunhai OS in the business application cross-processor architecture scenario. This is Yunhai OS, following the successful completion of the world's largest single cluster up to 1000 nodes cloud digital intelligence integration practice, and the establishment of "large-scale cloud cluster reference design specifications", once again leading the development trend of private cloud, established a cloud platform reference benchmark with "one cloud multi-core" as the core.
According to the data, SPEC is a third-party testing organization composed of dozens of well-known universities, research institutions, and IT companies around the world. It has a number of public standardized test specifications and models such as SPEC CPU, SPEC Power, SPEC Cloud, and SPEC ML. Among them, SPEC Cloud IaaS 2018 (SPEC Cloud for short), as an authoritative cloud computing performance benchmark, comprehensively evaluates cloud platform performance in terms of comprehensive performance, scalability, and instance deployment time. It is currently the most recognized IaaS cloud performance evaluation benchmark by enterprise customers in the world.
SPEC Cloud simulates IO-intensive and computing-intensive workloads in various cloud scenarios such as public cloud, private cloud, and hybrid cloud through YCSB and K-Means instances, so as to verify the comprehensive performance and scalability of the cloud platform. Among them, the YCSB task builds a Cassandra instance to simulate a standard social media NoSQL database application scenario. The test items include throughput, database insertion, read response time, and application instance configuration time; the K-Means task builds multiple instances through the big data benchmark tool HiBench to test the K-Means algorithm. The test items include completion time, Hadoop iteration count of instance running, and application instance configuration time. The entire test process evaluates the data plane performance, scalability and control plane performance of the cloud platform by continuously increasing the load pressure.
In this SPEC Cloud test, Inspur Yunhai deployed Yunhai OS on the x86+ARM hybrid processor architecture cluster, and achieved a number of indicators comparable to or even exceeding the performance of a single chip architecture cluster. Among them, in terms of average instance deployment time, Yunhai OS has achieved the best value in the SPEC Cloud IaaS 2018 cloud benchmark test, showing efficient and stable scheduling and concurrency capabilities on the control plane, and rapid pull-up and response capabilities of resource instances; in terms of performance scores, Yunhai OS’s performance score in the one-cloud multi-core scenario exceeds the SLA baseline by 20%, indicating that it still has excellent performance in the one-cloud multi-core scenario; In the multi-core scenario of one cloud, the business load can be effectively and elastically expanded.
The SPEC Cloud results show that Yunhai OS has high efficiency, high performance, and high scalability in a cloud-multiple-core scenario. It can realize hybrid deployment of cross-architecture resource pools, architecture-aware adaptive scheduling, rapid configuration and creation of application load instances, high concurrency and load balancing, etc. It can effectively help enterprise-level users complete unified management of multi-core resources in one cloud, load balancing scheduling, efficient and stable operation of applications, and elastic business expansion.
Advanced concept + hard core technology + prosperous ecology : Inspur Yunhai's "one cloud with multiple cores" problem-solving ideas
In Zhang Dong’s view, the reason why Inspur Yunhai has achieved such achievements lies in the support of advanced concepts + hard-core technology + ecology.
Advanced concept
Inspur Yunhai believes that the ultimate goal of one cloud with multiple cores is to support low-cost or free switching of user services between processors with different architectures. In terms of implementation, it can be divided into three steps of "Yes-Good-Excellent" to achieve the medium and long-term goal of one cloud with multiple cores.
- The first stage: "Have" stage - hybrid deployment, unified management, and unified view. At this stage, the cloud is used as the entrance, and the unified management of heterogeneous processor servers is realized through the cloud operating system, covering mainstream processors at home and abroad, providing a unified operation interface, operation mode, and unified monitoring and maintenance, etc., to solve the problem that the cloud resource pool can only carry servers of a single vendor or a single architecture processor, and the application cannot be deployed and coordinated across architectures.
- The second stage: the "good" stage - business traction, layered decoupling, and architecture upgrade. On the basis of the "have" stage , further realize low-cost cross-architecture switching of applications. Through the " layered decoupling " design of the resource layer, platform layer and application layer, the architecture upgrade is realized. At the resource level, rapid deployment of applications across architectures is achieved through cross-platform migration. At the platform level, cross-architecture hybrid deployment of platform applications is realized. For example, database clusters realize cross-architecture state synchronization and data synchronization on a cloud multi-core platform, and realize parallel operation of multiple instances on servers with different processor architectures. At the application level, cross-architecture traffic segmentation of applications can be realized through technologies such as service mesh.
- The third stage: "Excellent" stage - computing power standard, full-stack multi-core, ecological co-construction. From the perspective of business continuity,the ultimate goal of " one cloud with multiple cores " is to achieve a complete decoupling of applications and chip architectures. This requires the joint collaboration of the upstream and downstream of the hardware, cloud, and application industry chains to improve the quantification of effective computing power under heterogeneous architectures to support the equivalent switching of applications between processors with different architectures. Jointly promote the combination of applications and cloud infrastructure with one cloud and multiple cores, create a vertical integration solution, and realize cross-architecture non-aware switching at the application level. In terms of standards and evaluation, we will cooperate with professional software evaluation institutions and the upstream and downstream ecology of the industrial chain to form a one-cloud multi-core industry standard.
In terms of specific implementation, these three stages will also intersect. For example, Inspur Yunhai believes that it is in the "existing" stage, but it will also do the migration of resource packaging in the "good" stage, and even the standards and evaluation work in the "excellent" stage.
Core Technology
According to Qi Kaiyuan, Technical Director of Inspur Data Cloud Computing, and Liu Jian, Program Director, Inspur Yunhai's "one cloud with multiple cores" has achieved comprehensive technological leadership through continuous innovation.
- Open and Compatible: Compatible with a full range of processors, supports virtualization and container cloud management of X86, ARM, MIPS, Alpha and other architectures, and is the first to implement the unified pool management and scheduling of "one cloud with multiple cores". Inspur Yunhai is the first in the industry to realize the mixing of multiple chip resources in a single cluster, reducing the overhead of management nodes; it is the first to complete the adaptive scheduling function, which can automatically schedule the appropriate CPU type computing power according to business characteristics; it supports business traffic segmentation for multi-architecture chips, and can customize the copies and traffic split ratios of different architectures, and at the same time realize cross-architecture switching and migration of multi-core applications.
- Excellent performance: In terms of performance, Inspur Yunhai OS adopts X86+ARM hybrid processor architecture to complete the world's first SPEC Cloud benchmark test for a cloud-multi-core scenario. The performance score in a cloud-multi-core scenario exceeds the SLA baseline by 20%.
- Cloud-intelligence integration: Upgraded to the intelligent computing OS, which fully integrates general-purpose computing power, AI computing power, and HPC computing power, realizes the integrated scheduling of multiple computing power, and provides a rich full-stack service catalog.
"Prosperity" ecology
A prosperous ecology is indispensable for any technology and business development. "One cloud with multiple cores" is no exception. The ecology of "one cloud with multiple cores" is, in the final analysis, an ecology of a set of industrial chains with the cloud as the core and from top to bottom.
At present, the "one cloud multi-core" ecology is still in its infancy, and it is only at the level of chips and complete machines, and it is not "prosperous". Only when more vendors such as databases and middleware enter the ecology, from application vendors to platform vendors to the entire IT industry chain, support "one cloud with multiple cores", which is the sign of the true "prosperity" of the ecology. Only then will we have the confidence to say that we have built a new ecology of "one cloud, multiple cores" with Chinese characteristics that leads the world.
Inspur Yunhai is focusing on solving the global proposition of "one cloud with multiple cores" through the troika of "advanced concept + hard core technology + prosperous ecology", and has given its own solution ideas.