Ultimate efficiency is the foundation for the cloud native database tdsql-c to settle down

What kind of database is needed in the cloud native era ？ How to build a database service ？ Chengbin, technical director of Tencent cloud database, believes that , In the future, cloud database will be upgraded from hosting as the core to extreme efficiency , Help business reduce cost and increase efficiency . From the perspective of database management and Application , Cloud manufacturers 、 resources 、 Customer triangle There are three dimensions of efficiency behind it ： System efficiency 、 Operational efficiency 、 Business efficiency , When these efficiencies are at their best , Costs will fall sharply . How Tencent cloud database revolves around these three dimensions , To create the ultimate efficiency to enable the business ？

Brought in this period Chengbin, technical director of Tencent cloud database The sharing record at the main forum of the 4th cloud native industry conference ：

Video full ：《 Evolution of database technology in the cloud native era 》

Hello everyone , I am Chengbin, an engineer of Tencent cloud database , On behalf of the team, I would like to talk with you about the thinking on database technology in the cloud native era and the implementation practice of Tencent cloud in the cloud native database field .

Business background

We observed that , stay IT Architecture area , Cloud Nativity has replaced cloud computing , Ushered in a new stage . Inside Tencent , Each business system will look at its own cloud origin maturity score , Constantly optimize and improve . Cloud native compared to cloud computing , There are three changes . On application scenarios , From the pan Internet industry to a whole industry digitalization , Thus, the data scale is further increased . Infrastructure , It has also changed . On the cloud IaaS PaaS Become more mature , startup 、 The unicorn 、 Head listed companies and even applications related to people's livelihood , their IT The architecture is built on the cloud .

These changes caused our team to think , What kind of database is needed in the cloud native era ？ How do we build database services ？

We think , Cloud databases should be upgraded from hosting to extreme efficiency .

Before talking about ultimate efficiency , Let's take a look at the triangle behind the cloud database . The triangular relationship is that cloud vendors use and manage hardware resources , Provide services to customers , Customers use the resources behind services to conduct business activities . There are three efficiencies behind it , System efficiency 、 Operational efficiency and business efficiency .

first System efficiency , System efficiency refers to two levels of things , One level is to put CPU Memory 、 All the hard disk resources are used up , Don't idle ; Another level is , More requests can be processed with the same resources .

the second Operational efficiency Cloud manufacturers' online large-scale operation efficiency , Database tuning is common ,1w An example 1 individual DBA It can be done ,10w Instances or 100w An example ,DBA The number of people needs 10 individual 、100 A? ？

Third Business efficiency It refers to how to help businesses improve development efficiency . for instance , The game needs the same service all over the world , The level of business architects has been tested before , If the database can provide low latency global deployment capability , Can it greatly simplify the business architecture ？

Let's think about it , These three efficiencies point to the same word “ cost ”.

System efficiency

Next , I will share with you , How does Tencent cloud database create extreme efficiency to help businesses reduce costs and increase efficiency .

Traditional database architectures are designed based on resource integration , This design brings benefits in terms of delay , But lost flexibility and flexibility , It seriously affects the efficiency of the system .

We designed Cloud native architecture for computing and storage separation , Pool storage , Computing stateless , Greatly improve the utilization of system resources . Changes in architecture will certainly introduce many new problems , For example, the network delay problem of computing storage separation . In addition to introducing RDMA The network reduces the network delay . We also Innovative introduction of multi tier storage ： The L2 cache is added to the computing node , Cache the hot data locally , Reduce remote IO. On storage , In addition to the multi copy mechanism , We have also achieved Erasure code + Compression , Without reducing data reliability , Provides less than 1 Low cost storage capabilities for replicas .

Do a good job in system efficiency Be sure to ensure the efficiency of the architecture first . What is architectural efficiency ？ for instance , It is also a computing storage separation architecture , Insert a 1KB Record , How much data needs to be transferred to the back-end storage ？ Many products have to be transmitted 16KB, Magnified 15 times , And our architecture design only needs 1KB Just add a few bytes .

A lot of times , The magnitude brought about by the architecture gap, It's hard to compensate by software engineering . therefore , We especially pursue the ultimate efficiency in architecture , Because the product needs an excellent gene .

Operational efficiency

About operational efficiency , I mainly share two points with you , Business continuity and autonomy .

For business continuity , We proposed Panoramic business continuity solutions , Introduced Failure prediction mechanism To proactively deal with potential hazards in advance . On failover , We go through Cache preheating To achieve the lossless switching of the business .

In terms of fault diagnosis capability , Combined with customer load change and multi-level detection information , Generate Full link diagnostic report and Root cause analysis , Improve operation and maintenance efficiency and continuously find and optimize more abnormal scenarios .

Compared with traditional database architecture , We Reduce the time of business continuity failure scenario switching 74%.

With the rapid growth of business scale , Customer diversification , There are more and more cases of irrational use of databases , Database problem analysis and tuning have brought us great challenges .

How to solve this problem ？ We're on AI for system This road . adopt DB combination AI, use AI Instead of manual tuning 、 The diagnosis 、 Optimize .

Let's start with database parameter invocation , Through deep reinforcement learning, etc AI Algorithm No human intervention is required to realize parameter adjustment . On this basis , Through the precipitation of internal expert experience and the combination of various Al Algorithm , The tuning speed can be reduced by increasing the concurrency .

At present, we support thousands of parameter adjustment scenarios 、 The online application performance is improved at the highest level 235%, Tuning timeliness is reduced from a few hours to a few minutes , While saving manpower , It also greatly reduces the tuning time .

The above intelligent parameter adjustment results are also supported by the database SIGMOD Double employment , Click on extended reading 《 Three papers were selected for the international summit SIGMOD, Great Tencent cloud database 》.

Business efficiency

Business efficiency , We first thought of the original core of cloud primordial ： Ultimate flexibility , Pay as you go , No use, no pay . for instance , In the instance of the public cloud , There are quite a number of small and medium-sized long tail businesses , Such as periodic development testing 、 Personal blog 、 Low frequency data search, etc , Their workload on the database is not high , More sensitive to use costs .

For such users , Break the traditional product form with fixed specifications , Create the computing resources according to the business load Auto scale , Charge according to actual use , No use, no charge serverless form . In order to realize the automatic scaling of computing resources , The database is Start stop A lot of optimizations have been made in the aspect of , Ensure that the system can complete shutdown and restart recovery within a few seconds , Enhance the ultimate business experience .

At present, our database serverless The product exceeds only in wechat ecology 50W Applet developers provide database bases , Click on extended reading 《 Tencent cloud database cooperates with wechat cloud hosting , Help business reduce cost and increase efficiency 》.

We've found that socializing 、 game 、 Multinational enterprises represented by advertising have a big pain point when using databases , Businesses usually deploy applications in many places around the world , for example , The game goes out to sea , Players log in to accounts all over the world . To ensure the login experience , Multiple account databases will be deployed around the world , And realize the database synchronization between each other , This requires a lot of manpower and energy to maintain . Customers need a database that can be synchronized globally and accessed nearby , To simplify the application architecture .

In response to this question , We are about to launch the global database function , Provide cross Region Deployment capability , And provide access points everywhere , It is convenient for customers to read and write nearby . Through physical log replication technology , It solves the delay problem of cross region replication .

It is not enough to provide users with a good database engine , To improve business efficiency , It also needs to be provided for users Provide all kinds of convenient and easy-to-use SaaS Tools .

Tencent cloud native database provides a variety of SaaS It's a tool , for example , Database agents help businesses cope with traffic peaks with high performance , The database autonomy tool intelligent tuning helps customers improve the overall system level performance SQL performance , Database encryption 、 Back up 、 Auditing the full link helps secure enterprise data 、 More compliant .

stay Mixed load processing scenarios On , We also have an integrated design , Through a system , Provide... In a way that is transparent to the business OLTP and OLAP Two data query capabilities , The current gray level is medium , It will be online soon .

Future outlook

In the future, we will continue to explore The advanced nature of the structure 、 Extreme software engineering capability . meanwhile , We will deeply understand the customer's business scenarios , Put the architecture capability 、 Software engineering capabilities cover more business scenarios . such as , Explore the fusion of data models , One library for multiple purposes 、 Use unified framework to support more mixed load handling , To help businesses reduce the complexity of database architecture , So as to improve the application efficiency .

Again , We continue to practice AI for system, From plug-in intelligence to embedded intelligence , Let the database achieve the highest level of autonomy . We continue to make progress on the road of database with optimal efficiency .

That's all for today's sharing , Thank you again and the cloud native industry conference , thank you .