[translation] modern application load balancing with centralized control plane

author ： Armand - Sultantono 2022 year 3 month 10 Japan

The original for Armand Sultantono stay Snapt Blog Guest articles published on

The centralized control plane of application delivery allows you to use a single platform to meet all your application delivery needs . It brings together teams involved in the application lifecycle , These teams are usually scattered across many systems and platforms , There are still improvements .

The application development team is authorized to deploy the application itself , Adopt self-service . Platform operation （NetOps、SecOps、Site Reliability etc. ） You can set up highways instead of roadblocks , Ensure compliance with corporate safety standards , Without killing the speed of development and deployment .

What is? " Plane ", What are data plane and control plane ？

Let's first discuss the basics .

These days ," Plane "、" Data plane " and " Control plane " These words are frequently used . In different technical fields , These terms may have different connotations , But they all adhere to the same basic pattern .

• One Plane It is an abstract representation of where some operations occur .
• Data plane （ Sometimes called forwarding plane ） It's a systematic " Worker " Where it is . In the network , The data plane is responsible for the actual processing of communication , Until the packet is forwarded from one interface to another .
• Control plane It is the management layer that manages the data plane , For data plane " Worker " Set rules , And provide users with a management interface . In a decoupled model （ We'll talk about this next ）, The control plane manages the structure of the system , Distributed data plane . In the network , Control plane control and configure the team of data plane workers , To properly manage traffic .

Why should we decouple the control plane from the data plane ？

Decouple the data plane from the control plane , The idea of allowing networks to be programmed is an interesting concept , It is software that defines everything （SDE） And infrastructure, code （DevOps） The basis of . This is consistent with the modernization of software system , To decouple applications from the infrastructure they run .

Generally speaking , The decoupling of data and control plane has the following benefits .

• Operational efficiency -- Be able to abstract the complexity of specific components , And manage the whole complex system in a more effective way than managing independent and isolated systems .
• flexibility - Flexibility to allocate and reallocate resources more effectively ： Automatically scale infrastructure and mix best platforms .
• Cost optimization - By leveraging ubiquitous connectivity as commodity hardware and ready to provide services , Achieve efficiency and cost savings .

Traditional integration of modern applications ADC The problem of

In application delivery , image Load Balancer and Network application firewall （WAF） Such components traditionally have an integrated control plane and data plane -- The user interface and logic reside in the same application as packet processing and forwarding .

background

The tradition of load balancers at the edge of data centers ADC The pattern can be traced back to the load balancer （ And most network functions ） When hardware is dedicated .

The traditional load balancer is a single deployment in a high availability configuration , Usually Take the initiative / Passive pair （ One of them is free ）, It can handle all or most of the application combinations of an organization . Traditionally , The load balancer is quite static , Its basic settings rarely need to be modified . They work in all applications running behind them , Provide security 、 Reliability and performance .

today , Load balancer is a comprehensive application delivery platform , So they are also called " Application delivery controller " or ADC.ADC Responsible for broader and deeper application centric functions , This is far more dynamic and application specific than low-level network functions . This includes TLS/SSL uninstall 、 Application level DDoS Attack and zombie mitigation 、 Network applications and API Protect 、 Access management and Network content optimization - Just a few examples. .

problem

stay There are hundreds of applications In the environment of , Conventional ADC The model becomes a workload " bottleneck ", And may produce a big " Explosion radius ", Thus endangering your business .

Besides , Conventional ADC The way is right Allowing the rapid introduction of new functions A limitation of agile development methods . This is especially true when applications are deployed as microservices , Because microservice applications are not single applications that need to be maintained and updated as before , It's made up of smaller Services , These services all require specific application delivery requirements , And updated by different application development teams at a higher speed . With each new release of application code , Often need to be right ADC Make various modifications , Such as modifying routing rules 、 Security strategy, etc .

Traditional technologies that support modern applications ADC The model will lead to continuous strategic changes , This may seriously affect ADC, Destroy the stability of your application delivery and security . This also makes it necessary to update the service in the specified maintenance window Platform operation The team brought pressure , Because they have to manage the delivery of many services , The burden is heavy .

Decentralized every application ADC The problem of

Conventional ADC Is stifling development , And concentrate the workload on a single bottleneck . There is no way to solve this problem ？ Is it possible to use every application ADC Or for each tenant ADC To solve this problem ？

Yes , Build a by application or by tenant （ Customer ） Separately managed ADC The model of is completely feasible , Because we are no longer subject to expensive special hardware equipment . All modern ADC Both provide virtual and even containerized forms , In a small space （ Low computing and storage requirements ） The inner package has high performance and rich functions .

advantage

This dispersed ADC Model Reduced bottleneck and explosion radius . It allows for every application " Adjust reasonably "ADC Of scale , as well as Authorize the application team At your own pace Iteration and release New code .

for example , application A Of ADC Clusters may contain more ADC example , Run in the application B More powerful virtual machine specifications , Because it requires larger system requirements to support heavier workloads . Besides , application B Our application development team can push new functions and modify their ADC Strategy , Without worrying about affecting the application A Service for , Because they are independent clusters .

problem

When you use this method to empower the application team , If you have established a system controlled by many teams ADC Instance fleet , Things can easily get horribly out of control . Facing distributed deployment , Make sure that Compliance 、 Governance and consistency and Not simple .

for example , Applications A Suppliers may be used X Of ADC, And the app B（ Does not meet any safety standards ） The same may be used ADC, But previous versions of the code were unsafe , Or maybe a supplier Y Completely different ADC, And all of a sudden, you've joined the spread of tools .

How to combine integrated and decentralized control planes ADC Benefits of the model

The enterprise solution is a centralized control plane for application delivery . The centralized control plane provides a platform , It provides a " A single glass window ", Keep it all ADC Deployed inventory , And allow you to deploy in one place 、 Configure and operate your entire application delivery infrastructure .

In this mode , In the data plane ADC Examples and logic and management in the central control plane UI It's decoupled .

ADC The deployment of can be logically organized in a reasonable way -- If necessary , It can be by application and by tenant . Through this arrangement , We can get " The best of both worlds " The effect of . The platform operation team can provide ADC Instance and set " guardrail " Or global policy , And enable application developers to manage their own applications ADC Strategy .

Of course , If the control plane supports role-based access control （RBAC）, Let different characters use the control plane for their specific purposes , It is possible . therefore , In this model , We bring together all the groups involved in the application lifecycle . The Internet 、 Security 、 Application developers 、 Website reliability engineer and so on . A single platform , It also simplifies the spread of heterogeneous tools , And minimize the risks and costs associated with operational burden and risk exposure .

When can I use the centralized control plane for application delivery ？

When you manage applications that are massively scalable and distributed , And there are many teams involved in the application lifecycle , The centralized control plane is a necessary platform .

There are several key scenarios , You may want to use a centralized control plane .

• You have a mix and Microservice architecture , You need a simple way to manage your application delivery in both environments .
• You have one Cloudy strategy , And hope to get the benefits of cost optimization , But I hope to unify cross platform tools and observation capabilities .
• You have Application of distribution in edge calculation , And a simple method is needed to manage a large number of edge nodes in a consistent way .
• You want to speed up the deployment of applications , Empower application developers without compromising security .
• You have accumulated many tools , Perhaps there are several control plane solutions to manage a subset of tools .
• You want to have an internal or SaaS Flexibility of the solution , To help solve the above situation .

What are the benefits of a centralized control plane for application delivery ？

Decoupling and centralized control plane are prerequisites for the modernization of application delivery . A centralized control plane platform can let you .

• Centrally manage application delivery for hybrid and microservice architectures , Whether in the enterprise or in the cloud .
• Control through load balancing and application delivery （ADC）、API management 、 Network applications and API Protect （WAAP）、 Threat Intelligence And other key functions , Tools for integrated application delivery are distributed .
• It's simpler 、 Manage your deployment in a more flexible way , From modern times, based on Web Of GUI To allow centralized data output and input and integration into third-party tools API-- Whether it's a cloud platform for automatic expansion API It is also used for continuous integration and continuous deployment （CI/CD） Of API drive control .
• Improve the overall application delivery infrastructure Visibility and Analysis Ability , Be able to extract summarized insights and detailed indicators .
• Provide self-service function , Allow control to be delegated to the application team . Once other teams （ for example SecOps and NetOps） Guardrails are set , Application developers can quickly 、 Independently configure and customize their own end-to-end application delivery infrastructure , In most cases, the infrastructure is code （IaC） Method .

The next generation of intelligent control plane for modern application services

If your application infrastructure is designed 、 Monitoring and management are simplified by the control plane , Then the next generation control plane will add intelligence to this equation .

It also uses machine learning （ML） The implementation of adaptive and automatic network control promotes the programmability of your data plane -- Faster than a team of human experts 、 More precise .

The next generation control plane is also ready for emerging applications of edge computing and cloud native architecture .

Expectations for the next generation of control planes .

• API Priority control plane , Sure " Headless " function , And connect to other structures dedicated to the management plane , Not as a terminal control plane .
• advanced ML Ability , To eliminate human intervention and errors -- Through adaptive traffic rules and self-healing 、 Self expanding infrastructure for dynamic traffic patterns 、 React to anomalies and threats .
• Community driven Threat Intelligence , Use historical and predicted traffic as well as internal and external （ Public and private ） Real time data collected by the source , Prevent future attacks in advance .
• Intention based policies allow users to describe their requirements in the most basic terms , To define results and high-level operational goals , Without listing conditions and actions , Even become an expert in every discrete technology function . If application developers want to " Follow the sun " Dynamic deployment of applications , The control plane will instantiate and configure the required number of instances on multiple cloud platforms , To meet the traffic demand of specific regions , Without user intervention .
• Highly optimized communication protocol between control plane and data plane , It can remotely control millions of devices in real time and in parallel , It has near zero delay and extremely high scale and elasticity .

Conclusion

The centralized control plane for application delivery will help accelerate application deployment and simplify life cycle management .

This is the spread of management tools 、 The solution of island operation , And by addressing the challenges of hybrid and multi cloud deployment , Bring everyone involved in the application lifecycle under one roof .

Please have a look at Snapt Nova, This is a Centralized control plane Next generation application delivery platform .Nova It can be deployed inside the enterprise or turn key SaaS.Nova It's an arbitrary cloud 、 Multi location 、 Centralized platform , Used to deploy 、 Control and monitor the large-scale delivery of your applications .Nova The built-in ML Automated and simple intent based templates and configuration files simplify agile application delivery on a large scale .