Sogang University team's storage system tech accepted at top HPC conference SC26

Researchers at Sogang University have developed an innovative object storage system named OASIS, which promises to dramatically reduce data movement within high-performance computing environments. Led by Professor Kim Young-jae of the Computer Science department, the team's work addresses a critical bottleneck in large-scale scientific data analysis: the time and resources consumed by transferring massive datasets from storage to analysis servers.

The OASIS system introduces a technique called hierarchical query execution. Instead of simply offloading simple operations like filtering to external servers, OASIS intelligently determines the optimal storage layer within the system itself to perform computations. This approach minimizes data transfer not only out of the storage device but also between different internal tiers, such as from disk enclosures to storage servers.

According to the research, OASIS can reduce internal data movement by as much as 99.998%. In practical evaluations using scientific workloads, internal data transfer decreased from 20GB to just 76KB. Furthermore, the system achieved a reduction in end-to-end latency by up to 25.1% compared to existing state-of-the-art systems.

“

The utility of storage-side computation has been consistently raised, but the situation where storage system internal resources were not sufficiently utilized has persisted. Our team redefined this as a placement problem of 'at which layer within the storage should execution be terminated,' aiming to actually utilize the hierarchical resources from enclosures to storage servers. As a result, we were able to solve the bottleneck of internal data movement that had not been revealed until now.

The research, titled 'OASIS: Hierarchical Query Execution over Multi-Layer Computation-Enabled Object Storage,' was conducted in collaboration with SK Hynix Memory System Research Center and the U.S. Los Alamos National Laboratory. It has been accepted as a full paper at SC26, the 38th International Conference for High Performance Computing, Networking, Storage, and Analysis, a top-tier global academic conference with an acceptance rate of 19.4% this year.

Beyond its academic significance in redefining storage-side computations as a placement optimization problem, OASIS offers practical industrial benefits. It can be implemented using existing storage hardware without requiring specialized equipment and maintains compatibility with standard S3 interfaces, allowing seamless integration with current data analysis engines. The reduction in data movement also translates to lower energy consumption, a crucial factor for the sustainability of increasingly power-hungry AI and data infrastructure.

“

I am grateful to Professor Kim Young-jae for his guidance, and to the researchers at SK Hynix and Los Alamos National Laboratory for their collaboration.