Distributed Metro eResearch infrastructure with secure and native long-haul RDMA over standard Telco fibres
Mr Ashrut Ambastha1, Mr Rhod Brown1, Mr Sudarshan Ramachandran1
Higher education and research institutions are geographically distributed across multiple campuses either in different regions of a city or across multiple states. It is not always feasible to create a centralised infrastructure that contains all compute and storage resources in one single location.
By applying the learnings of creating large datacenter computational units, we are proposing to extend these concepts to the edge and build an architecture called MetroPOD to interconnect a fleet of citywide compute units and storage using Remote Direct Memory Access (RDMA).
Utilising existing fibre deployments, MetroPOD will allow any institute to build an ultra-low latency, in-network computing fabric for AI and HPC that stretches across the metropolitan area. Data generated at a given site can be natively mounted over RDMA to any participating remote site within the metro limits thereby accelerating processing time. The main building block is NVIDIA MetroX-2, a native InfiniBand solution scalable up to 40km with standard optics
Another challenge in this scenario is achieving zero trust security. NVIDIA BlueField data processing unit (DPU) provides a secure and accelerated infrastructure for any workload in any environment, from cloud to datacenter to edge by offloading, accelerating, and isolating a broad range of advanced networking, storage, and security services.
In this session we will cover all of the above aspects and provide a blue print that can serve as a starting point for a MetroPOD design and deployment.
Ashrut Ambastha is a Principal Engineer at NVIDIA working on networking fabric for HPC clusters and AI datacenters. He is also a member of application engineering team that works on product designs with networking silicon devices. He holds an MSc and MTech in Electrical Engineering from Indian Institute of Technology-Bombay.