Who is using VeloDB?

52TT is a VoIP social platform for online game players to hang out. With over 200 million registered users, it has established official partnership with five eSports leagues, including LPL, KPL, PEL, IVL, and CDL.

How do they use VeloDB?

Based on VeloDB, 52TT built their own user profiling and behavioral analysis platform. Data sources of this platform include data that generated by user behaviors and daily operation, and the tags assign to them. The source data is ingested into the data platform to create user profiles, on which basis analysts divide the users into groups, extract insights about users, and evaluate effectiveness of their business strategies. The results they derive will work as reference for their operation, A/B testing, or be used in their business systems.

How are they benefiting from VeloDB?

The reason that 52TT settled on VeloDB as their choice of analytic database was that the old ClickHouse-based data processing architecture was unable to fulfill some of their needs as their business grew:

• The real-time tags were updated frequently, while the limited data updating functionality of ClickHouse made that inconvenient.
• OOM errors happened a lot and interrupted service processes.
• There was no transaction support, thus no DDL atomicity guarantee.
• Under the ClickHouse architecture, to handle a large data size, each node has to be mounted with SSD, which is not only costly but also arduous for cluster scaling and maintenance.
• As ClickHouse falls short in multi-table joins, most data needs to be arranged in flat tables before they are ready for analysis and queries.

VeloDB brings some fresh air to the picture:

• Partial column update: VeloDB supports partial column update, so any changes to the tags can be updated easily and flexibly in real time. It also supports metadata changes in a lightweight and atomic way. Thus, it can make sure that the tags are always up-to-date.
• Efficient memory management: VeloDB manages memory by its MemTracker mechanism. Any query execution errors will be identified immediately and the relevant process will be killed. This will ensure stable functioning of query services and prevent interruption to business.
• Storage-compute separation: VeloDB provides an out-of-the-box cloud solution that separates storage and computing, so it can help reduce storage costs and maintenance efforts.
• Easy multi-table joins: VeloDB supports not only flat table queries but also complicated cross-table joins. Joining a 10-billion-row table with a billion-row tag table only takes 3~5 seconds. This avoids the trouble of flat table making and enables more possibilities in data analysis.

From ClickHouse to VeloDB, 52TT saves 40% of its costs in server resources and increases efficiency in user segmentation by 10 times. By the elastic scaling, multi-cluster deployment, and resource isolation capabilities of VeloDB, they also witness a 5~10 times increase in their service availability. 52TT is going to include more tasks in the VeloDB data platform, such as risk control and real-time user segmentation. Inspired by the successful use cases of LLM-powered OLAP with VeloDB, they are also looking into building a more intelligent user profiling services by integrating VeloDB and Large Language Models.