Voltaire Ltd. (Voltaire) has introduced a new 40 Gb/s InfiniBand director-class switch, the Grid Director 4700, that provides low latency. This new switch and the company’s Unified Fabric Manager software offer high performance, scalability and ease-of-use for large, high-performance clusters. The new switch features 324 ports of 40 Gb/s InfiniBand connectivity, with the option to double the capacity to 648 ports utilizing double-density fabric boards.
When the switch and software are used together, they not only provide high performance anticipated from InfiniBand, but also enterprise-class reliability, high availability and the service-level manageability, based on the company’s leadership in bringing the high-performance computing into commercial applications and the data centers.
The Voltaire Grid Director 4700 is part of the company’s fourth generation switch family. The double-density fabric boards are basis for HyperScale architecture, a new stackable architecture for building larger configurations into hundreds and thousands of nodes, with lower latency and greater simplicity than the alternative solutions. The 19U high Grid Director 4700 has below 300 nanosecond port-to-port latency to accelerate the performance of applications running on server and the storage scale-out fabrics. The switch runs a smart distributed software architecture that offers the best signal quality, improved performance optimization, and better resiliency that are all key challenges in large-scale deployments. Moreover, its low 6.5-watts-per-port power consumption makes switch a greener choice for the energy-efficient data centers.
The company started shipping a 36-port version of 40 Gb/s switch, the Grid Director 4036, in December 2008. “We’ve already deployed Voltaire’s 40 Gb/s InfiniBand edge switch and are experiencing improved application performance for our Applications Research and Development cluster, which translates into solving more complex problems in less time,” said Goldi Misra, Head of High Performance Computing Solutions Group, C-DAC, Pune. “A high-port count switch leveraging 40 Gb/s InfiniBand such as the Grid Director 4700 will give us increased bandwidth and throughput to capitalize on the processing power of multi-core servers and scale up to thousands of nodes.”
Advanced Management Software Further Improves Performance:
The company has also introduced Unified Fabric Manager software, the comprehensive management software to optimize the performance of large server and storage scale-out fabrics. The new Unified Fabric Manager software eliminates the additional latency that is created in complex environments where multiple applications rely on single fabric to simultaneously transmit the server traffic and access storage. According to the company’s benchmarks, the new Unified Fabric Manager software improves latency by as much as 10X in real world environments that run multiple applications and the traffic types.
Compatible with the company’s 20 Gb/s and 40 Gb/s switching platforms, including the Grid Director 4700, Unified Fabric Manager software monitors, analyzes and optimizes fabric performance to enhance the efficiency and utilization of large InfiniBand fabrics. Unlike other management software platforms that are device oriented and involve tedious manual processes, the new Unified Fabric Manager software offers IT managers with a logical view of their infrastructure. This bridges the traditional gap between servers, applications, and fabric elements, creating more effective and business oriented way to manage and scale out high-performance fabrics.
The new Unified Fabric Manager software efficiently monitors and analyzes the fabric bottlenecks and errors, enabling IT managers to quickly identify failures, inefficiencies and performance issues and then take corrective action. The new Unified Fabric Manager software includes a new fabric optimization technology that matches fabric resource allocation and configurations to specific application requirements leading to reduced congestion, lower latencies, and increased application performance.
“Large fabrics bring a tremendous amount of complexity. Management software that automatically discovers, virtualizes, monitors and optimizes the fabric infrastructure as well as accelerates active applications is a valuable asset to our clusters that scale into the hundreds of nodes,” said Professor Rudolf Lohner, Karlsruhe Institute for Technology (KIT). “Having a centralized view across all of the infrastructure gives us more control over our large-scale environments.”
The new Unified Fabric Manager software integrates easily in the existing environments and has a rich web services API and SDK for easy integration with third-party management platforms.
“Voltaire continues to lead the high-performance server and I/O connectivity market with the availability of the highest performing, high density solutions that deliver 40 Gb/s performance, low latency and scalability,” said Asaf Somekh, vice president of marketing, Voltaire. “The combination of the Grid Director 4700 and Unified Fabric Manager software not only delivers the best performance for large clusters, but also provides the most management features to simplify scale-out.”
The new Grid Director 4700 and the new Unified Fabric Manager software optimizes the I/O throughput for multi-core servers, improving speed of high-performance computing applications in the industries including manufacturing, energy, financial services, life sciences, government, and research and education, leading to greater productivity, improved service levels and the higher revenues.
HP’s multi-core Cluster Platforms are among those compatible with company’s new Grid Director 4700 switch and the new Unified Fabric Manager software.
“By leveraging high-performance computing solutions to accelerate business processes such as research, engineering or analysis, customers can improve decision making, allowing them to grow their businesses,” said Ed Turkel, manager of product marketing, Scalable Computing and Infrastructure organization, HP. “Together with Voltaire, HP Cluster Platforms significantly increase performance and simplify management for large, complex systems.”