SC19: Liqid Showcased Software Disaggregation for Converged Servers

At SC19, Liqid, Inc. announced its composable infrastructure software can disaggregate traditional, off-the-shelf converged servers into pools of shareable data center resources.

The Command Center software enables underutilized, static devices such as GPU, NVMe, FPGA accelerators, and Intel Corp.‘s Optane memory technology to be redeployed for adaptive use and improved utilization. These resources can be shared widely as required across all major fabric types, including PCIe Gen 3, PCIe Gen 4, Ethernet, and IB. Unified multi-fabric support from the company delivers a wide reach for composability and increases infrastructure elasticity for established converged and hyperconverged environments.

The firm has showcased its composable infrastructure solutions at the 2019 International Conference for HPC (SC19), conducting collaborative demonstrations with Nvidia Corp., Western Digital Corp., One Stop Systems, and other industry IT firms.

“Liqid’s sophisticated, multi-fabric composable infrastructure software enables comprehensive, adaptive management and automation of previously fixed data center resources without upending existing data center environments,” said Sumit Puri, CEO and 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 HPC applications that are increasingly driven by AI.“

Disaggregates hyperconverged hegemony at SC19
With limited ability to disaggregate data center resources, traditional converged and hyperconverged systems can become uneven, with some resources sitting idle while others are taxed to their limits. The company’s composable infrastructure solutions address these limitations through software, permitting organizations to scale converged and hyperconverged architectures and right-size for all levels of compute activity, down to bare metal.

At SC19, the company showcased:

Demo with Western Digital Corp.: Booth-to-booth composable NVMe-oF via Ethernet;

Demo with One Stop Systems: The ‘Honey Badger’ (LQD4500), the firm’s composable NVMe add-in-card capable of four million IO/s of random performance, over 24GB/s of throughput, and less than 20ms, available in capacities up to 32TB;
Software-defined, bare-metal composable servers built with disaggregated hardware resources, including Nvidia Corp.‘s GPUs, NVMe storage, FPGA data accelerators, 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 Corp. and Mellanox Technologies, Ltd.;
Liqid high-memory solutions with up to 12TB of Optane memory and 184TB of NVMe storage in a 2U server, enabling in-memory calculations for 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 center,” said Paresh Kharya, director, product marketing, accelerated computing, Nvidia. “Liqid’s software-defined composable infrastructure solutions can help customers adapt to the evolving demands of AI, ML, virtualized environments, 5G/edge deployments.“

Read also:
$28 Million Series B Funding for Liqid, Total at $50 Million
Doubled Q/Q revenue with record profit.
November 14, 2019 | Press Release
FMS: From Liqid, Element LQD4500 PCIe Add-In-Card NVMe Up to 32TB
Up to 4 million IO/s of random performance, more than 24GB/s throughput, and transactional latency of 20µs
August 14, 2019 | Press Release