Today Mellanox announced that EDR 100Gb/s InfiniBand solutions have demonstrated from 30 to 250 percent higher HPC applications performance versus Omni-Path. These performance tests were conducted at end-user installations and Mellanox benchmarking and research center, and covered a variety of HPC application segments including automotive, climate research, chemistry, bioscience, genomics and more.
Dan Olds from OrionX.net presented this talk at the Switzerland HPC Conference. “Dan Olds will present recent research into the history of High Performance Interconnects (HPI), the current state of the HPI market, where HPIs are going in the future, and how customers should evaluate HPI options today. This will be a highly informative and interactive session.”
“This talk will focus on challenges in designing programming models and runtime environments for Exascale systems with millions of processors and accelerators to support various programming models. We will focus on MPI+X (PGAS – OpenSHMEM/UPC/CAF/UPC++, OpenMP, and CUDA) programming models by taking into account support for multi-core systems (KNL and OpenPower), high-performance networks, GPGPUs (including GPUDirect RDMA), and energy-awareness.”
In this video, Dana Brunson from Oklahoma State describes the mission of the Oklahoma High Performance Computing Center. Formed in 2007, the HPCC facilitates computational and data-intensive research across a wide variety of disciplines by providing students, faculty and staff with cyberinfrastructure resources, cloud services, education and training, bioinformatics assistance, proposal support and collaboration.
Ira Weiny from Intel presented this talk at the OpenFabrics Workshop. “Individual node configuration when managing 1000s or 10s of thousands of nodes in a cluster can be a daunting challenge. Two key daemons are now part of the rdma-core package which aid the management of individual nodes in a large fabric: IBACM and rdma-ndd.”
“Infiniband Virtualization allows a single Channel Adapter to present multiple transport endpoints that share the same physical port. To software, these endpoints are exposed as independent Virtual HCAs (VHCAs), and thus may be assigned to different software entities, such as VMs. VHCAs are visible to Subnet Management, and are managed just like physical HCAs. We will cover the Virtualization model, management, addressing modes, and discuss deployment considerations.”
Xiaoyi Lu from Ohio State University presented this talk at the Open Fabrics Workshop. “Single Root I/O Virtualization (SR-IOV) technology has been steadily gaining momentum for high performance interconnects such as InfiniBand. SR-IOV can deliver near native performance but lacks locality-aware communication support. This talk presents an efficient approach to building HPC clouds based on MVAPICH2 and RDMA-Hadoop with SR-IOV.”
“Using RDMA, NVMe over Fabrics (NVMe-oF) provides the high BW and low-latency characteristics of NVMe to remote devices. Moreover, these performance traits are delivered with negligible CPU overhead as the bulk of the data transfer is conducted by RDMA. In this session, we present an overview of NVMe-oF and its implementation in Linux. We point out the main design choices and evaluate NVMe-oF performance for both Infiniband and RoCE fabrics.”
Todd Rimmer from Intel presented this talk at the OpenFabrics Workshop. “Intel Omni-Path was first released in early 2016. Omni-Path host and management software is all open sourced. This session will provide an overview of Omni-Path including some of the technical capabilities and performance results as well as some recent industry results.”
Yuval Degani from Mellanox presented this talk at the OpenFabrics Workshop. “In this talk, we present a Java-based, RDMA network layer for Apache Spark. The implementation optimized both the RPC and the Shuffle mechanisms for RDMA. Initial benchmarking shows up to 25% improvement for Spark Applications.”