SK Planet's Oh!RaBang Cuts Cloud Costs by 50% Through Serverless Migration and Bundle Optimization
Key point
To handle the unique traffic patterns of live commerce, SK Planet's Oh!RaBang migrated to a serverless architecture and optimized its frontend bundle, cutting cloud costs by about 50%.
Details
SK Planet's live commerce platform Oh!RaBang has an unusual traffic pattern where traffic surges sharply within 1-2 minutes of a broadcast starting. In the existing ECS and ELB-based structure, manual scaling using Lambda was applied to prevent service outages caused by Auto Scale delays, but this led to wasted computing resources and unpredictable cost increases.
Transition to a Serverless Architecture
The team's analysis found that the main cost drivers were network traffic and computing resources. Accordingly, they migrated to a serverless structure based on CloudFront and S3. Since this approach has minimal storage costs and is billed per request, it eliminated the idle instance costs that were the biggest problem with the existing method. In addition, CloudFront offered cheaper network costs than ELB for handling the same amount of traffic, improving efficiency.
Frontend Bundle Optimization
For additional cost savings, the team also worked to reduce network usage on the frontend server. To resolve the large bundle size caused by structural limitations of the existing Next.js environment, they carried out a migration to Vite. This reduced the JavaScript bundle size from 1.4MB to 0.4MB, about a 70% reduction, and also improved dev server speed and build performance.
Results
Through the serverless migration and bundle optimization, the team achieved cost savings of over $5,000 per month, equivalent to about 50% of total cloud costs. During pre-verification, they confirmed that CloudFront guarantees up to 250,000 TPS, ensuring stability even during traffic surges.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.