Uploaded November 2025 | Updated September 2026, 2 weeks ago
Enhancing Ceph for Infinite Scalability at LINE - Jinmyeong Lee & Sungjoon Koh, LINE Plus
At LINE, user-exchanged media is stored in Ceph via RADOS Gateway. As the volume of data grew significantly, a single Ceph cluster could no longer meet our demands for scalability. To address this, we built a new Hyper Scale Object Storage system that combines infinitely scalable structures with multiple Ceph clusters and other cost-effective S3-compatible storage, all served through a single endpoint.
We focused on three key engineering solutions to build the system. First, to integrate multiple backends seamlessly, we introduced an abstraction layer that handles bucket distribution and request routing, while natively supporting inter-cluster APIs like CopyObject. Second, for efficient and secure data tiering, we implemented a migration API which enables transferring SSE-C encrypted objects between Ceph and heterogeneous S3-compatible storage without decryption. Additionally, we implemented AES256-CTR for compatibility. Lastly, to automate the entire migration process, we developed a Migrator, which schedules and manages data migration tasks.
We will share how we enhanced Ceph to meet the unique needs of Hyper Scale Object Storage in high-traffic, large-capacity environments.
Enhancing Ceph for Infinite Scalability at LINE - Jinmyeong Lee & Sungjoon Koh, LINE Plus
At LINE, user-exchanged media is stored in Ceph via RADOS Gateway. As the volume of data grew significantly, a single Ceph cluster could no longer meet our demands for scalability. To address this, we built a new Hyper Scale Object Storage system that combines infinitely scalable structures with multiple Ceph clusters and other cost-effective S3-compatible storage, all served through a single endpoint.
We focused on three key engineering solutions to build the system. First, to integrate multiple backends seamlessly, we introduced an abstraction layer that handles bucket distribution and request routing, while natively supporting inter-cluster APIs like CopyObject. Second, for efficient and secure data tiering, we implemented a migration API which enables transferring SSE-C encrypted objects between Ceph and heterogeneous S3-compatible storage without decryption. Additionally, we implemented AES256-CTR for compatibility. Lastly, to automate the entire migration process, we developed a Migrator, which schedules and manages data migration tasks.
We will share how we enhanced Ceph to meet the unique needs of Hyper Scale Object Storage in high-traffic, large-capacity environments.
