Meta's AI Training and Inference Infrastructure is growing exponentially, presenting a dramatic scaling challenge. The company needs to build and evolve its network infrastructure to connect numerous training accelerators and ensure the network runs smoothly with stringent performance and availability requirements for RDMA workloads.
Requirements
- Experience with using communication libraries, such as MPI, NCCL, and UCX.
- Experience with developing, evaluating and debugging host networking protocols such as RDMA.
- Experience with triaging performance issues in complex scale-out distributed applications.
- Understanding of RDMA congestion control mechanisms on IB and RoCE Networks.
- Experience in developing systems software in languages like C++.
Responsibilities
- Responsible for the overall performance of the communication system, including performance benchmarking, monitoring and troubleshooting production issues.
- Identify potential performance issues across the stack: comms lib, RDMA transport, host networking, scheduling and network fabric.
- Develop and deploy innovative solutions to address the performance issues.
- Active member of a multi-disciplinary team to develop solutions for large scale training systems.
Other
- Bachelor's degree in Computer Science, Computer Engineering, relevant technical field, or equivalent practical experience.
- BS/MS/PhD in relevant fields (EE, CS), with 4+ years work experience.
- Understanding of AI training workloads and demands they exert on networks.
- Understanding of the latest artificial intelligence (AI) technologies.
- Experience with machine learning frameworks such as PyTorch and TensorFlow.
