Liqid showcases software disaggregation for converged servers
19 Nov 2019 Denver - Liqid's composable infrastructure software can disaggregate traditional, off-the-shelf converged servers into pools of shareable data center resources. Liqid Command Center software enables underutilized, static devices such as GPU, NVMe, FPGA accelerators, and Intel Optane memory technology to be immediately redeployed for adaptive use and significantly improved utilization. These powerful resources can be shared widely as required across all major fabric types, including PCIe Gen 3, PCIe Gen 4, Ethernet, and Infiniband. Unified multi-fabric support from Liqid delivers the industry's widest possible reach for composability and dramatically increases infrastructure elasticity for established converged and hyperconverged environments. Liqid will showcase its composable infrastructure solutions at the 2019 International Conference for High Performance Computing (SC19), conducting collaborative demonstrations with NVIDIA, Western Digital Corp., One Stop Systems, and other industry-leading IT innovators.
"Liqid's sophisticated, multi-fabric composable infrastructure software enables comprehensive, adaptive management and automation of previously fixed data centre resources without upending existing data centre environments", stated Sumit Puri, CEO & Cofounder, Liqid. "By unlocking traditional converged hardware with Liqid Command Center, IT administrators can optimize their overall footprint substantially and build out disaggregated infrastructure to more effectively prepare for the challenges associated with evolving high performance computing applications that are increasingly driven by artificial intelligence."
With limited ability to disaggregate data centre resources, traditional converged and hyperconverged systems can quickly become uneven, with some resources sitting idle while others are taxed to their limits. Liqid composable infrastructure solutions address these limitations through software, permitting organisations to scale converged and hyperconverged architectures and right-size for all levels of compute activity, down to bare metal.
At SC19, Liqid is showcasing the following:
- Demo with Western Digital (booth #1275): Booth-to-booth composable NVMe-over-Fiber (NVMe-oF) via Ethernet;
- Demo with One Stop Systems (booth #1935): The "Honey Badger" (LQD4500), Liqid's composable NVMe add-in-card capable of 4 million IOPS of random performance, over 24GB/s of throughput, and less than 20ms, available in capacities as high as 32TB;
- Software-defined, bare-metal composable servers built with disaggregated hardware resources, including NVIDIA GPUs, NVMe storage, FPGA data accelerators, Intel Optane memory technology, CPU, and networking devices;
- Software powered disaggregation of traditional converged servers into independent pools of composable resources and GPU-over-Fabric (GPU-oF) in collaboration with NVIDIA and Mellanox Technologies;
- Liqid high-memory solutions with up to 12 TB of Intel Optane memory and 184TB of NVMe storage in a single 2U server, enabling in-memory calculations for extremely large datasets.
"Multi-fabric, multi-GPU, and GPU-oF computing models are increasingly being adopted as techniques to pool and share and scale NVIDIA GPU compute, and play an increasingly important role in the modern data centre", stated Paresh Kharya, director of product marketing for Accelerated Computing at NVIDIA. "Liqid's software-defined composable infrastructure solutions can help customers adapt to the evolving demands of artificial intelligence, machine learning, virtualized environments, 5G/edge deployments."