Challenges

  • Unstable databases

    Most Meitu databases consist of a primary node and multiple secondary nodes. The nodes balanced loads based on DNS but could not eliminate hot spots or load imbalance issues. Databases were unable to automatically perform full backup, full restoration, incremental synchronization, or rate limiting, thus could support few service scenarios.

    Most Meitu databases consist of a primary node and multiple secondary nodes. The nodes balanced loads based on DNS but could not eliminate hot spots or load imbalance issues. Databases were unable to automatically perform full backup, full restoration, incremental synchronization, or rate limiting, thus could support few service scenarios.

  • Poorly utilized big data clusters

    Big data accounted for about half of Meitu's data and was increasing rapidly. Meitu required different amounts of compute and storage resources, but the resources were on the same physical machines and could not be decoupled for elastic scaling. Data had multiple copies and is frequently transferred, occupying much storage space and bandwidth.

    Big data accounted for about half of Meitu's data and was increasing rapidly. Meitu required different amounts of compute and storage resources, but the resources were on the same physical machines and could not be decoupled for elastic scaling. Data had multiple copies and is frequently transferred, occupying much storage space and bandwidth.

  • Heavy O&M workload and costs

    Each database administrator in Meitu maintained more than 100 databases and instances on average, struggling with tiresome and repetitive maintenance work. Meitu used Hadoop for big data analysis and paid a high price to fix bugs, improve performance, and scale the system.

    Each database administrator in Meitu maintained more than 100 databases and instances on average, struggling with tiresome and repetitive maintenance work. Meitu used Hadoop for big data analysis and paid a high price to fix bugs, improve performance, and scale the system.

Solutions

  • Smooth migration to a cloud-native platform

    Meitu and Huawei signed a strategic cooperation agreement. Meitu upgraded its infrastructure to cloud-native infrastructure, greatly improving service performance and resource scalability. Huawei Cloud designed a solution that used cloud databases and decoupled compute and storage resources, smoothly migrated Meitu's seven major service modules to Huawei Cloud, and achieved significantly higher database performance and stability.

  • Strong capabilities and diverse DR solutions

    Huawei Cloud Relational Database Service (RDS) uses a distributed HA architecture, and is highly stable and reliable. RDS splits read and write requests, distributes requests, and balances loads to enhance database performance. Routine O&M was automated on the database management platform to greatly reduce DBAs' workloads. A range of DR solutions were tailored for migration, synchronization, subscription, and disaster recovery (DR) scenarios to guarantee data security.

  • Decoupled storage and compute on the cloud big data platform

    The Huawei Cloud BigData Pro solution helped Meitu use the same cloud-native infrastructure for AI and big data. Compute and storage resources were completely decoupled and could be separately scaled. The MapReduce Service (MRS) big data service provided cost-effective multi-core CPUs, and was fully compatible with the open-source ecosystem. Meitu's services were migrated to the cloud without modifying code or interrupting services. Data using different protocols was all stored in an Object Storage Service (OBS) data lake, thus was widely accessible and does not need many copies.

Benefits

Higher productivity and security

  • Higher performance

    The cloud native infrastructure achieves high performance and scalability, able to guarantee database stability even in high concurrency scenarios. QPS is increased by 3.5 times. Meitu can efficiently handle massive service requests of more than 2 billion users in unexpected peak hours.

  • Lower costs and higher quality

    Real-time online services and offline compute services are centrally managed on the cloud native infrastructure platform and can be easily maintained. The database O&M efficiency is improved by 70%, and the average time for accessing core service databases is reduced by about 67%. On-demand scaling of compute and storage resources achieves 40% higher resource utilization and 30% lower costs.

  • Security and reliability

    Multiple DR modes are available for Meitu to choose. The backup reliability reaches 99.999999999%. Terabytes of data can be restored within minutes.