Backend.AI utilizes a proprietary GPU-centric orchestrator and scheduler to ensure optimal resource placement and multi-node workloads distribution for AI and high-performance computing. Additionally, it incorporates a storage proxy to parallelize data I/O, further enhancing its efficiency in managing computing resources and unlocking their maximum potential.
Distributed processing of multiple containers using multiple nodes
- Built-in scheduler-wide multi-container resource placement and relocation
- Supports high-speed networking technology such as RDMA
GPU optimization technology
- Implementation of CUDA Optimal Solutions based on NVIDIA Partnership
- Industry's only container-based multi/partial GPU sharing (fractional GPU™ scaling)
Enforce container resource constraints in HPC libraries (e.g. BLAS)
- Powerful system resource control through libc overriding (CPU core count correction, etc.)
Dynamic sandboxing: programmable and rewriteable syscall filter
- Supports rich programmable policies compared to apparmor/seccomp