Ali yunyili: how does yunyuansheng solve the problem of reducing costs and improving efficiency?

author ： Yili

2020  Since then , COVID-19 has changed the operation of the global economy and people's lives. . Digital production and lifestyle have become the new normal in the post epidemic era . today , Cloud computing has become the digital economic infrastructure of the whole society , Cloud native technology is profoundly changing the way enterprises use and use the cloud . There are many ways to use cloud native technology to help enterprises achieve cost reduction and efficiency increase  IT  Topics concerned by the person in charge .

Alibaba has always been an explorer and practitioner in the original field of cloud . There are several main lines in the cloud native path of Alibaba group ：

• The first stage is the Internet application architecture .2007  From the year onwards , With the explosion of Internet traffic , Ali started to build an Internet scale distributed application architecture , Self research micro service 、 news 、 Distributed database and other core middleware .

• The second stage is the cloud of infrastructure . Alibaba cloud  2009  In, he began to develop Feitian cloud operating system , The business of the group began to go up to the cloud . meanwhile ,2011  In, Alibaba group began to explore the technology of landing containers , Accelerate application migration to the cloud , Maximize the elasticity of the cloud , Calculate the cost by optimizing the off-line mixing part , here we are  2019  In, the core system was fully put into the cloud .

On this basis , Alibaba Group has started to realize a comprehensive cloud native upgrade . We insist on open source technology 、 Alibaba cloud products and group applications   The trinity . To  2021  year , Realized business  100%  On the cloud , application  100%  Yunyuan biochemistry .

Cloud native technology has brought huge value dividends to Alibaba . At present, Alibaba Group has the largest in the world  K8s  colony , A single cluster has more than 10000 nodes , It can uniformly support e-commerce 、 Search for 、 big data 、AI  And other diversified applications .2021  Ali Shuang  11  The calculation cost of promoting the peak value has decreased compared with last year  50%.Serverless  Also landed in a large number of scenes , R & D efficiency is improved  40%.

Thanks to Alibaba's cloud native large-scale practice , Alibaba cloud has built an advanced platform for enterprises 、 Pratt Whitney's cloud native product family , At the same time, it serves Alibaba group and customers in various industries .2022  year  1  quarter , In the authoritative consulting organization  Forrester  In the public cloud container platform analyst report released , Alibaba cloud container service  ACK  Become comparable  Google  The global leader of , This is also the first time that a Chinese technology company has entered the container service leader quadrant .

Problems faced by enterprises on the road of cost reduction and efficiency increase

Over the past few years , The cloud native technology represented by containers is developing rapidly , In the latest  CNCF  Developer survey , stay  2021  In the third quarter of the year, there were more than  700  Wanyun native Developer . It is the consensus of most developers that using cloud native technology can drive enterprises to achieve cost reduction and efficiency increase . But what we can also see is , stay  2021  year  CNCF《FinOps Kubernetes Report》 According to the Research Report of ,68%  Percent of the respondents said that their company had been in the past year  Kubernetes  The cost of computing resources for the environment has increased . What is the reason behind this ？

Through communication and analysis with enterprises , We found that enterprises are currently facing five major problems ：

• Difficult planning .
    When the business is migrated to the container scenario , Capacity planning for applications is required , Over allocation of resources can lead to waste of resources , Oversold resources will lead to stability problems .

• Difficult to charge .
    Compared with traditional applications, container applications are more flexible and dynamic , Resources can be created and released on demand , This also brings greater challenges to cost estimation .

• It is difficult to divide the accounts .
    Different from traditional application deployment and resource binding . Now multiple container applications share one  K8s  colony . Multiple can run on a single compute node  Pod, and  Pod  It can stretch , Dynamically migrate between nodes . Application layer and resource layer metering and billing in space 、 Many dimensions such as time can not achieve one-to-one correspondence , Cause the complexity of cost management .

• Difficult to optimize .
    Cloud native technologies such as ： elastic 、 Mixing part 、Serverless、 Oversold and other technologies have their own typical scenarios . If not used properly , For example, the elastic configuration is wrong , It may lead to unexpected waste of resources and even stability problems .

• Difficult to manage .
    Hybrid cloud has become an enterprise  IT  The new normal of architecture .Kubernetes  It can help enterprises shield basic differences . However, the financial management ability varies in different environments , Lack of unified and open consumption data model for management , It makes it difficult for enterprises to conduct overall cost analysis and optimization from the overall perspective .

In recent years, , With the acceleration of Enterprise Cloud , Cloud financial management （FinOps） The concept of is mentioned and adopted by more and more enterprises ,FinOps  It is a cloud operation mode , Instead of just empowering leaders 、 Best practices and culture are combined , To improve the organization's ability to understand cloud costs . This is an approach that brings financial responsibility to cloud spending , Enable the team to make informed business decisions .FinOps  Enhanced  IT、 engineering 、 financial 、 Procurement and collaboration between enterprises . It makes  IT  Can develop into a service organization that focuses on using cloud technology to add value to business . When cloud native technology and  FinOps  Concepts are intertwined , It gave birth to cloud primordial IT Cost management （Cloud Native FinOps） Idea , It is  FinOps  An evolution and evolution of the concept in the cloud native scene .

In order to solve the new challenges brought by Cloud Architecture and cloud native technology , Enterprises begin to pay attention to new cost management methods . adopt IT、 financial 、 Business and other teams work together , While ensuring business development, it helps enterprises obtain better financial control and predictability .

Cloud native enterprise  IT  Cost management plan ： Accelerate the business  FinOps  process

Alibaba cloud combines the practice of industry finance integration with  FinOps  idea , Provides native product capabilities , It provides enterprises with full link financial operation and management guarantee on cloud . Alibaba cloud container service launched the enterprise cloud native service  IT  Cost management plan , Help enterprises in the scene of cloud native cloud , Provide enterprises  IT  Cost management 、 Enterprises  IT  Cost visualization 、 Enterprises  IT  Cost optimization and other functions .

5  month  19  Japan , By China Academy of information and communication （ hereinafter referred to as “ China Academy of communications ”） Co sponsored with China Communication Standardization Association “2022  Cloud management and cloud Network Conference ” Successfully held . It was released at the meeting 《 Credible cloud • Cloud cost optimization tool capability requirements  -  The first  1  part   Native tools 》 Standards and the first batch of evaluation results . Alibaba cloud relies on its product capability of cost management on the cloud , Passed all with full marks  33  Capability indicators , Become the first cloud service provider in China to pass the evaluation .

Alibaba cloud Enterprise Cloud native  IT  The cost management scheme has five core functions ：

Core function one ： Unique cloud native container scene cost allocation and estimation model .

  In order to solve the problem of inconsistent life cycle between business unit and billing unit in container scenario , Container service puts forward a unique cost estimation model combining billing and measurement , And add expense strategy （ Payment type 、 Saving plan 、 vouchers 、 User discount 、 Bidding fluctuation ）、 Allocation factor （CPU、 Memory 、GPU  card 、GPU  Video memory, etc ）、 Resource form （ECS\ECI\HPC） And other factors , Implement for  Pod  Cost estimation of dimensions and cost allocation of cluster proportion . All resource costs of the cluster in one stage are aggregated through bill analysis , Combined with  Pod  The cost allocation capability of dimension realizes a complete cost allocation and estimation model of cloud native container scenario .

Core function 2 ： Multidimensional cost insight 、 Trend prediction 、 Root cause tripping .

  Support clusters 、 Namespace 、 Node pool 、 application （label  Wildcard match ） Four dimensions of cost insight , The cluster dimension focuses on the distribution of cloud resources 、 Trends in resource costs 、 The ratio of cluster water level to waste and the trend and prediction of cluster cost , Can help IT The administrator accurately judges the trend of cost consumption , Prevent scenarios that exceed the budget ; Namespace focuses on cost sharing , Support short-term cost estimation and long-term cost allocation , Support dispatching water level 、 Resource usage 、 Correlation analysis of cost trends , Assist the Department Administrator in cost estimation , RIH analysis cost waste , Improve departmental resource utilization ; The node pool dimension focuses on resource cost planning and governance , By instance type 、 When the unit is checked 、 Dispatch water level 、 Correlation analysis of utilization rate and water level , assist  IT  Asset managers optimize resource mix and payment strategies . application （label  Wildcard match ） Dimension focuses on cost optimization in the field scenario , for example ： big data 、AI、 Work offline 、 Online applications and other upper application scenarios , Real time cost estimation and task level cost accounting can be carried out through the cost insight of the application dimension .

Cost insight through four dimensions , The cost optimization function and solution of the whole scene can be supported by data , Carry out cost reduction and efficiency increase with reasonable reasons .

Core function 3 ： Cost optimization capability of the whole scene 、 Solution coverage .

  For the actual business scenarios of different enterprises , Alibaba cloud container service provides the establishment of resource portraits for the whole scene 、 Cost optimization capabilities and solutions . Besides , The optimization strategy of enterprises for cost , Most of them need to be supported by business scenarios , There will be customization and secondary development in many scenarios . therefore , Alibaba cloud container services are native to the enterprise cloud  IT  The cost insight provided by the cost management scheme is completely decoupled from the upper optimization scheme , Cost insight can be achieved through four dimensions , Measurement and evaluation of cost optimization means covering the whole scene .

Core function 4 ： Multi cluster / cloudy / Hybrid cloud full type cloud cost management capability .

  Cloudy is a new trend for enterprises to go to the cloud , The billing models of different cloud vendors are quite different , for example ： The common annual and monthly payment method of domestic cloud service providers 、 Common credit card withholding of international cloud service providers / after payment 、 Some cloud service providers support saving plans and reserved instances, etc . All these provide more challenges to the cost analysis ability of multi cloud cloud cloud management plane . Alibaba cloud container services are native to the enterprise cloud  IT  The cost management scheme is realized by providing unified billing and Inquiry Access and default of cloud service providers , Support mainstream cloud service providers 、IDC  Access of cost data of self built computer room . And cost management is carried out through a consistent cloud native container scene cost allocation and estimation model . Cooperate with enterprise cloud native distributed cloud container platform  ACK One, Not only does it provide multiple clusters 、 Unified cluster management for multiple environments 、 Unified resource scheduling 、 Unified data disaster recovery and unified application delivery capability , It also provides a unified financial management capability .

Core function 5 ： Enterprise Cloud native IT Expert services in cost management .

  Enterprise Cloud native  IT  Cost management is not just a product capability or solution , It is also an enterprise in the cloud original era IT management 、 Organization process 、 The evolution of culture . Alibaba cloud container service team and Alibaba cloud space-based team , Provide complete services through Alibaba cloud asset manager  FinOps  Products and expert services covered by the concept .

such as , We can use multi-dimensional cost analysis 、 Insight function , Understand the cost and resource utilization of the application . And can provide trend prediction . Provide decision-making basis for enterprise financial management . In order to meet the diversified governance needs of enterprises , We also provide an open data model , Can pass  Prometheus、OpenAPI  And other methods are integrated into the enterprise's own governance process . With cost insight , Let's see what means we can use to achieve cost optimization .

Diversified flexibility ： Examples of flexible containers can be found in  30  Second expansion  3000 Pod

Elasticity is one of the core capabilities of the cloud , It can effectively reduce the computing cost .ACK  In the resource layer and application layer, it provides rich elastic strategies .

At the resource level , When cluster resources are insufficient ,ACK  Clusters can take advantage of  cluster-autoscaler  Automatically create new node instances in the node pool . According to the application load , choice  ECS  virtual machine , Examples of DPCA bare metal , super-popular . Based on Alibaba cloud's powerful elastic computing power , We can achieve 1000 node capacity expansion at the minute level .

stay  ACK  A more simplified solution in a cluster is to use  ECI  Elastic container instance to achieve elasticity .ECI  Based on lightweight virtual machines  Serverless  The container running environment , Strong isolation 、 High elasticity , No operation and maintenance 、 Features of capacity free planning . Examples of flexible containers can be found in  30  Second expansion  3000 Pod, Can easily deal with unexpected news events , Or support batch computing services such as autopilot simulation .

It is worth mentioning that , We can use  ECS  perhaps  ECI  Bidding instance of , It can use Alibaba cloud's idle computing resources , Cost discounts can be as low as for pay as you go instances  90%. Bidding examples are very suitable for stateless and fault-tolerant applications , Such as batch data processing or video rendering . At application level ,Kubernetes  Provides  HPA  By  Pod  The horizontal expansion of , and  VPA  Conduct  Pod  The vertical expansion of .ACK  Built in machine learning based  AHPA  programme 、 To further simplify the elastic experience , Improve the elasticity of  SLA.

K8s  Built in level  Pod  Automatic telescopic （HPA） There are two shortcomings ：

The first is the hysteresis of elasticity , The elastic strategy is based on the passive response to the monitoring indicators , In addition, since the application itself starts 、 Preheating also takes some time , In the process of expansion , Business stability may be affected ;

The second is the complexity of the configuration ,HPA  The running effect of depends on the configuration of elastic threshold . Too aggressive configuration may affect the stability of the application , The configuration is too conservative , The effect of cost optimization is greatly reduced . It takes trial and error to reach a reasonable level . And as the business changes , There will also be a need to readjust the flexibility strategy .

Alibaba cloud cooperates with the Dharma Institute , Launched  AHPA, According to the historical resources , Predict the elastic cycle and consumption , Expand capacity in advance to ensure service quality , Already in the rookie  PaaS  platform 、 Alibaba cloud intelligent voice service has been verified in many scenarios . Help intelligent semantic interaction products realize  90%  Before the business comes  Ready,CPU  Increased utilization  10%  save  20%  The cost of resources .

Upgrade the capability of mixed department ： stay  K8s  Provides an enhancement to the scheduling capability on the

With the wide application of cloud native technology ,Kubernetes  The workload of the above computing types is getting richer and richer , We can schedule through reasonable arrangement , Make full use of peak shaving and valley filling effect between loads , Make the workload more stable 、 More efficient 、 Use resources in a lower cost way . This is what the industry often mentions  “ Mixing part ” Concept .

Alibaba is in  2011  Began to explore container technology , And in  2016  In, the technology research and development of the mixed department was started , So far, it has undergone several rounds of technical architecture upgrading , Eventually evolved to today's cloud native hybrid system architecture , It has realized a cloud native hybrid department with a full-service scale of more than 10 million cores , Mixed part day average  CPU  Utilization rate exceeds  50%, Help Alibaba save a lot of resource costs .

The mixed department is a cost control core built with heavy money within Internet enterprises , It condenses many thinking and optimization experiences of business abstraction and resource management , Therefore, it usually takes years of grinding practice for the mixing part to gradually stabilize and produce production value . however , Does every enterprise need a high threshold to use the mixed department , It takes a lot of investment to produce value ？

Based on the practical experience of super large-scale production within Alibaba group , Alibaba cloud recently opened the cloud native hybrid project  Koordinator, It aims to create the lowest access cost for users in the cloud native scenario 、 The best solution for mixing efficiency , Help users and enterprises realize the continuous dividend release after the cloud is native . It's in  K8s  It provides the enhancement of the scheduling capability , Including three core competencies ：

• Differentiation  SLO guarantee ：
    stay  Kubernetes  On top of that, it abstracts a set of  QoS  Resource scheduling mechanism , For example, delay sensitive online tasks , and  Best effort  Type preemptible computing tasks . In the popularization of improving resource utilization , Let low priority tasks , Impact on delay sensitive tasks  < 5%;

• Fine resource scheduling ：
    Include  CPU、GPU  Topology awareness 、 Reserve resources 、 Interactive preemption 、 Defragmentation 、 Resource portraits 、 Hot spot dispersal and other fine scheduling capabilities ;

• Task scheduling ：
    Big data and  AI  Related task scheduling , such as  Gang、 Batch 、 Priority preemption and elasticity  Quota（ Inter queue borrowing ） etc. , So as to better apply the entire cluster resources .

Koordinator  The project is fully compatible with upstream standards  K8s, No intrusive modifications are required . Alibaba cloud container service provides product support , Users can also apply in their own scenarios based on open source projects . so to speak ,Koordinator  The open source , More enterprises can see and use the capabilities of the cloud native hybrid department , Help enterprises accelerate the process of cloud original biochemistry . On the technical ,Koordinator  It can help enterprises realize more load access to  Kubernetes  platform , Workload rich scheduling container , Then it gives play to the characteristics of workload peak staggering and time-sharing , So as to achieve efficiency 、 Cost benefit , Maintain a healthy form of long-term sustainable development .Koordinator  The project is still in the process of rapid development , Welcome to build .