Channel Corporation Reduces Migration Time by 96% via DynamoDB Table Separation and AWS Managed Services

Channel Corporation undertook architectural modernization to separate the Badge feature from the Amazon DynamoDB User table of its all-in-one AI messenger 'Channel Talk'. The existing User table held 1.8TB of data with approximately 1.68 billion records, serving multiple roles including profile and Badge. A surge in Badge traffic during One-Time Message (OTM) sending caused GSI Hotspot and Back Pressure, leading to overall service performance degradation.

Performance Stabilization via Table Separation

Since Badge data only requires lookups based on userId, GSI is unnecessary. Separating the Badge feature into a dedicated UserBadge table fundamentally eliminates the GSI Back Pressure issue. Additionally, because Badge traffic follows unpredictable spike patterns, On-demand capacity mode was applied, while the User table maintained Auto Scaling enabled Provisioned mode to optimize costs. After separation, Badge update conflicts decreased from 10,000 per minute to 80, and WCU usage on the User table was reduced by approximately 12–13%.

Migration Pipeline Innovation

The previous migration based on a custom Java application took approximately 128 hours to scan 1.8TB of data and placed load on the production table. Channel Corporation built a serverless pipeline combining DynamoDB Export to S3, AWS Glue, and Import from S3. This approach consumes zero RCU/WCU from the production table, eliminating throttling risks. Migration time was reduced by 96%, from 128 hours to approximately 5 hours and 30 minutes, and costs were cut by 36%, from $394.13 to $253.74.