The "China Artificial Intelligence Large Model Map Research Report" released by the New Generation Artificial Intelligence Development Research Center of the Ministry of Science and Technology shows that my country has released 79 large models with parameters exceeding 1 billion, and the "Battle of 100 Models" is about to begin. Large models not only bring new scenarios and new business formats, but also place extremely high requirements on computing power, which must not only be fast but also diversified. It is a general trend to provide different computing power for different applications such as AI training, AI reasoning, and numerical simulation, and to achieve full coverage of multiple computing types.
In fact, the large model storm is just the "tip of the iceberg" of the computing power revolution. With the gradual deepening of enterprise digital transformation and the rapid rise of cloud-native applications, the diversification of application scenarios has directly promoted the diversification and upgrading of computing power, from virtualization, cloud hosts, bare metal to containers to serverless... cloud foundation The upgrading and upgrading of architecture faces unprecedented challenges.
Should you continue to tolerate the generational upgrade method, or directly embrace the evolvable model? This question has become as serious and worthy of careful consideration as Hamlet's "To be or not to be".
Taking into account history and facing the future
Looking back at the evolution of IT infrastructure, from the complete split of the hardware monolithic architecture in the 1990s, to the subsequent virtualization of the x86 architecture, and then to cloud computing, before cloud native, the computing power on the cloud was basically relatively single. . After that, with the rapid rise of cloud native represented by containers, Kubernetes, etc., it was like opening a floodgate for computing power. Diversified and emerging computing power emerged in an endless stream, including bare metal, virtual machines, containers, Computing power such as Serverless, FaaS, and large models coexist, intertwine, and evolve together.
The diversification of computing power development is mainly reflected in the diversification of computing power needs stimulated by the diversification of computing scenarios, the diversification of computing power sources, and the diversification of computing power brands... Due to a qualitative leap, supporting diversified computing power on the public cloud can It is said that it comes naturally without any effort. However, for private clouds that evolve relatively slowly, diversification of computing power is still a high threshold for many users, and smooth upgrade is the first difficulty.
For example, the online banking we used on PCs was developed based on microservice architecture, but now more and more people choose mobile banking, which means that the application architecture needs to be rewritten. If you were given a choice, whether to tinker with the original microservice architecture or simply use a new cloud-native architecture for development, what would you choose? Although they know that cloud native is the most ideal way, many people may still hesitate. Why?
Is it really possible to get rid of the burden of history easily? Traditionally, IT infrastructure is built in a top-down manner, matching computing, storage, network and other resources according to application requirements. In this way, the architecture or system built at a specific point in time and under specific application requirements is destined to be a combination of resources and Applications are tightly bound to each "island".
Isolated islands are a reality. As long as the traditional system construction, deployment and application methods remain unchanged, the isolated islands will continue to exist, and they will accumulate more and more until they are difficult to return to. In the past, when the type and quantity of workloads were relatively small, this "1-to-1" construction method of applications and systems could relatively quickly meet the needs of users for launching new applications. However, as workloads become more and more complex, The more users there are, the shorter the window period for applications to go online, and there will be frequent gaps in users' resource supply capabilities and manpower. Moreover, in the face of isolated islands, management and unified operations have become very complicated, making it troublesome to improve efficiency and reduce costs. According to the traditional method, the expansion of the system will look relatively simple on the surface, that is, one system is simply superimposed on another. But in the later stage, when the number of systems becomes unbearable, it will be difficult to change the underlying architecture and the pain will be even more intense.
For many users, system expansion and upgrade has never been easy. Historically, every change in technical architecture has been a "cutover", which is equivalent to starting over from scratch, including the transformation from monolithic architecture to distributed, from distributed to microservices, and from ON CLOUD to IN CLOUD. exception. This will have a significant impact on the development of user services and must be properly handled.
The cutover change from mainframe to x86 is what we call a generational upgrade, but after all, it is a "makeover" in hardware. Although a generational upgrade is a helpless move, there seems to be no other better option. Method. However, in subsequent upgrades from virtualization to cloud to cloud native, the hardware has been "unified" and the software has the possibility of continuous and smooth upgrades that can evolve. If we continue to use the generational upgrade method, it will be tantamount to going back. .
In the context of diversified computing power, it is self-evident whether the upgrade of cloud infrastructure is generational or based on a sustainable and evolving cloud platform. In today's era, with increasing market uncertainty, the evolvable cloud platform actually provides users with a deterministic choice. You may not know how new technologies will change in the future, or how many "tricks" there will be in computing power. The evolvable cloud platform that "takes into account history and faces the future" can allow you to take it calmly and keep moving forward.
Unified platform + balanced design
Whether it is the old virtualization vendor VMware or the new private cloud expert EasyStack, they all have the same idea on the evolution of cloud native. They first build a cloud native base and then continue to add services on top. In order to meet the needs of diversified computing power, the cloud computing infrastructure must be able to achieve the unification of application, management and operation and maintenance at the user level, and at the product level, it must be able to unify functions, scenarios and quality control. More importantly, Achieve unification of development, testing, and iterative upgrades.
Not every manufacturer can easily meet the above-mentioned "three major unifications" requirements. Many manufacturers can only provide corresponding computing platforms for different scenarios in a superficial way, but do not have the ability to design a balanced architecture. As a result, the core issue of evolution is compromised or cannot be realized at all.
Take EasyStack as an example. Before it proposed an evolvable digital infrastructure, it also experienced struggles, constant trial and error, and continuous evolution. Before today's evolvable architecture took shape, EasyStack had only major core-level upgrades four times. From hardware to virtualization to containers, this is a gradual process. When the EasyStack platform was in the V4 and V5 versions, it initially only supported virtualization. Now it not only covers CPU, GPU, NPU, etc., but also provides encapsulation interfaces to better support various cloud services on it. More importantly, in this evolution process, EasyStack continues to input and integrate the capabilities of open source technology into existing commercial platforms through engineering methods, providing stability, reliability, and security for diverse computing power. of support.
EasyStack's new generation of evolvable cloud platform can not only support the operation of bare metal, cloud hosts, Kubernetes container clusters and multiple containers at the same time, but also provide capacity-based or high-performance cloud storage and unified cloud storage suitable for different computing power. Multi-VPC SDN network, as well as unified monitoring, logging and operation and maintenance services. This is the balanced design of the architecture, that is, at the infrastructure level, various capabilities such as computing, storage, monitoring, network, and security must not only be complete, but also balanced and matched, and connected with each other, without any "shortcomings", and It can be combined and adapted as needed to meet current and future expansion needs. It is conceivable that it is difficult to achieve a balanced design for a single computing power implementation architecture. With diversified computing power, the workload and complexity will increase exponentially.
Aiming at diverse computing power, the key to EasyStack's new generation of evolvable cloud platform that can be easily used lies in the separate design of its platform and cloud services. The lightweight unified digital native engine EOS, combined with diversified cloud services, can flexibly meet the needs of private deployment from small to medium-sized to ultra-large-scale clouds, whether it is virtualization, bare metal computing power, or cloud native , AI computing power needs can be seamlessly expanded to respond, and there is no need to purchase multiple sets of platforms and multiple sets of software.
Today, "one cloud, multiple cores" has become a basic feature of Xinchuang Cloud. One-cloud multi-core itself is a typical example of diversified computing power. As the name suggests, one cloud and multiple cores are based on a unified cloud and support chips of different architectures. Extending this "1-to-many" model, "1" represents an evolvable platform and cloud-native base, while "many" represents diversified computing power. What EasyStack strives to build is this "1" evolvable platform.
The starting point for evolution is “whole-house intelligence”
For example, generational upgrade is like the process from rough house to fine decoration when people renovate their houses; while incremental upgrade is like continuously adding some features with basic decoration on the basis of basic decoration. Home equipment with intelligent functions may be limited by the decoration style of the original house during the process of adding it. Compared with the evolution of IT systems, the IT infrastructure remains unchanged, with just "piecemeal improvements" on it; but it can evolve The best upgrade method starts with whole-house intelligence. No matter any new changes in the future, it can be easily integrated into the overall atmosphere and functions to achieve continuous upgrading of intelligence.
After the above comparison, you may have a better idea of which upgrade method is most appropriate and wise under what circumstances. What an evolvable architecture needs to do is make up for the gaps caused by the out-of-sync development of modern applications and IT infrastructure. For infrastructure platform manufacturers, support can evolve. It is not enough to only have a single technical strength. They must first solve the problem of "full" resource elements. They must have a complete technology stack and achieve a balanced design. At the same time, they must also Maintain openness and achieve complete decoupling of the infrastructure, so that the infrastructure can continue to evolve and be advanced.
Whenever new technologies and new requirements emerge, they can be easily supported and managed without changing the architecture and platform. This is the core value of the evolvable architecture. As a cloud operating system manufacturer, EasyStack provides a unified set of APIs with a purely software-defined solution. Based on the concept and architecture of "small core + multiple components", it realizes programmability and flexible combination of all scenarios, fully explaining what is The evolvability of cloud platforms. EasyStack has always focused on infrastructure, that is, the infrastructure platform and its associated basic capabilities, including diversified computing power, as well as cloud storage, cloud network, API, security, operating system, automation capabilities, etc.
For Chinese enterprise customers, the continuous increase of new computing power is the norm, and the long-term coexistence of steady-state business and sensitive business is a practical need. Only an evolvable architecture can effectively cope with the explosion of diversified computing power.
past/issue/return/review
Private cloud goes from "available" to "easy to use" and "evolvable" is a ruler.
Turning to enterprise-level basic platform software, let’s talk about the big picture of EasyStack
AWS+EasyStack hybrid cloud, the eve of the blowout of "public cloud + open source private cloud"