Abstract

The new energy vehicle sales market was extremely hot in December, especially for Li Auto, which won the top sales title in SUV category with its new L8/L9 and became the first "new vehicle force" with over 20 thousand deliveries in a single month.

What is the ideal winning strategy for such a bright "counterattack" performance? And how does the data-driven approach advocated by "Internet car makers" work? As the underlying infrastructure of data-driven, is the database really crucial for car companies?

In this article, we would like to explore these issues through the case study of YMatrix's collaboration with Li Auto.

01

In the just-past December, the new energy vehicle sales market was red-hot.

Although it is not the traditional sales season of the Golden September and Silver October, in 2023, the 13-year-old subsidy policy for the purchase of new energy vehicles will formally end, and a large number of wait-and-see demand will be released centrally before the policy closes, resulting in a wave of sales boom at the end of the year.

As a representation of traditional car companies, BYD continue to grow steadily, while the "new forces", which showed a slight weakness in the first three quarters, also made a turnaround in December.

NIO, XPeng and Li Auto have achieved a large growth in December, and Li is on behalf of the "new vehicle forces", for the first time to achieve the 20 thousand delivery volume.

What is more noteworthy is that, according to the data on the number of new insured cars, the two new cars, Li L8 and L9, have been crowned the king of the SUV market, beating a number of top tier big brands.

Many signs indicate that new energy has entered a new market stage. The market penetration rate of new energy vehicles has exceeded 27%, faster than the planned 25% pace of development while the growth rate is expected to slow down in 2023. The purchase subsidy has been formally ended, prompting more intense price competition. Lithium phosphate ore prices have risen all the way to push the price of batteries continue to go up, etc.

Uncertainty from both the market side and the production side are increasing, the industry will say goodbye to the collective surge. Major manufacturers need to do their utmost to improve efficiency reduce cost, and make innovation and breakthroughs in currently impossible areas,

Against this backdrop, Li Auto's recent turnaround performance is more remarkable.

02

The new car series L9 and L8 released in the middle of the year is undoubtedly an important point for Li Auto in 2022. In the fourth quarter, the L9 and L8 became the main delivery models, and it was at this stage that Li Auto not only reversed the continuous decline in deliveries in the first three quarters, but also set a new single-month delivery record, and set the stage for surpassing NIO in the whole year.

"Li L9, a family SUV with no rivals below 5 million yuan".

The tagline that triggered the hot topic of search engines is a straightforward illustration of Li AUto's differentiated positioning of "building cars for families".

I still remember driving home one day in the fall and seeing a new Li up ahead. Hadn't yet been identified which model it was, I saw a large screen hanging down from the center top through the rear window, which was playing a clip from Peppa Pig.

A large screen is not new at the moment, and what impresses me is the timing and location of this large screen, as well as the content that flashes on the screen.

On the way home, a family of three, each beautiful in its own way.

This precise capture and shaping of demand is built on the basis of extensive operational interactions. On the one hand, high-frequency interactions with users require operational means to maintain. On the other hand, operational data accumulated through long-term interactions are also needed to gain insights into the details of demand. More importantly, how to transform these data insights into standard processes in production? All these need to be realized through a set of digital operation system.

Thanks to its Internet gene, Li Auto has established a perfect digital operation system from the early stage. At the front end, there are various web-based business applications (e.g., App for car owners, management platform for after-sales service), which are connected to various types of business/data center through APIs for business empowerment and related resources. At the bback end lies a large-scale distributed and clustered infrastructure.

However, automobiles are more complex than the virtual Internet, involving more complicated business processes, denser offline physical assets, and more variable working conditions. Facing the complicated automotive industry system, how to design a set of digital operation system, Li Auto also faces many new challenges.

03

It is under such an opportunity that YMatrix and Li Auto met.

At that time, YMatrix proposed the concept of hyper-converged database, taking the time-series data scenario as a starting point, hoping to provide users with a set of unified data base that can be "stored with confidence" (high-performance writing and high-efficiency storage of multiple data types) and "used at will" (all-scenario data query and algorithmic analysis). The team of Li Auto's data platform, in turn, seeks a new technical solutions to comprehensively upgrade the business guarantee capability of the basic data platform.

The philosophical fit between the two parties was then translated into friction and collaboration all day and night. The project faced many challenges.

First of all, how to store a much larger amount of "big data", several orders of magnitude larger than Internet data. Ideally, thousands of data indicators should be recovered for each vehicle, and these data are constantly uploaded by a certain time frequency. As long as there are vehicles running, a large amount of data will be generated. Currently, the total amount of data doubles in basically a few months. In addition, unlike the relatively smooth operating environment of the Internet, vehicles run in a complex and changing real environment. Data collection and uploading will encounter a variety of extreme working conditions like data delay, disorder, loss, retrieval, additions, etc. It may occur at any time, and need to be able to flexibly cope with the situation.

Specifically to the underlying database, special attention needs to be paid to performance, storage efficiency and expansion convenience. Specifically, data write speed, support for complex write scenarios, storage compression ratio after data entry, tiered storage strategy are all worth paid attention. At the same time, in the face of fast-growing business needs, continuous expansion will be inevitable. Whether the expansion operation will affect normal business operation, how long it will affect, and whether the operation is simple and controllable are all questions that need to be answered.

Second, how to be more agile to respond to business needs, so as to empower the business, rather than becoming a development bottleneck. A typical scenario is: when the business encounters a specific problem, the business team will raise the demand, and need to quickly locate, analyze, attribute, and ultimately implement the solution; the analysis team will give the analysis scheme and algorithm, and the data team needs to be able to quickly provide the data interface to the fastest speed response.

In this case, the performance of the underlying database is the basic guarantee, all kinds of query performance must be the ultimate. In addition, the development difficulty should be low, the more simple and direct, the more standardized and common the database and the data interface is, the more sufficient monitoring and operation and maintenance protection will be, the more efficient the data development team will be.

After months of joint efforts around these challenges and requirements, the project finally went live in early 2022 and has been running smoothly ever since.

YMatrix replaced the 50-node OpenTSDB cluster with a 14-node cluster that carries the same size data, reducing cluster server usage by 2/3.

At the same time, YMatrix provides a set of smooth scaling solutions, which can realize the smooth expansion of cluster size without the interruption of business. The whole process can be operated entirely through the UI visualization interface, to provide operation and maintenance personnel with more simple, intuitive, process-oriented operation experience. Scaling is no longer an accidental journey, but a standard process to follow the SOP (standard process).

In addition, YMatrix also provides more than 100 core operations and maintenance monitoring indicators, covering write performance, query performance, cluster status and other aspects, providing rich monitoring data support for operation and maintenance personnel to accurately grasp the platform's operation status, as well as fault location and attribution analysis.

On the other hand, the original OpenTSDB platform does not support complex time-series queries such as aggregation query, window query, etc. Therefore, similar complex queries need to be realized through independent programming in Hive and Flink clusters, which has a certain technical threshold for code development and higher maintenance costs in the later stage. YMatrix natively supports comprehensive time-series query functions, such as aggregation query, window query, jump query, difference query, etc. Business developers can obtain query results directly through YMatrix by using standard SQL language, and the development time of indexes is dramatically reduced from several days to less than 1 hour. At the same time, YMatrix has also greatly improved query performance, such as indicator details and other common queries. Consumed time is reduced to less than 1 second, and the maximum reduction reaches more than 90%.

04

Back to our question at the beginning: can databases help new energy vehicle companies win the second half?

Of course not, but data can.

Making data the first factor of production is a philosophy shared by Li Auto and YMatrix.

Through vehicle operation data, Li Auto captures the most subtle user needs, drives business iteration, and continuously improves the efficiency of the established operation system. While on the production side, a huge amount of data on production lines and equipments is constantly being generated, and the intelligent manufacturing system constructed on the basis of these data will have unlimited space for imagination. This may be one of the elements for Ideal to win the fierce competition in the future.

YMatirx hopes that through its own technology and code, the data in various scenarios can find their own optimal "habitat", create a data base, stimulate the power of data convergence, and help enterprises produce data more efficiently, use data more simply, and mine data more deeply.

The year 2023 has come and gone, and the uncertainty has not completely dissipated in the post-epidemic era. New energy vehicles will usher in changes, and more industries will choose to respond to the changes through digital transformation, and the concept of letting data become the first factor of production will take root in more enterprises. YMatrix hopes to work with these enterprises to explore the road of industrial digitization together.