Dynamic Infrastructure for Modern Workloads

How to get to 90% utilization in the data center.



The 44th IT Press Tour had the opportunity to meet with Sumit Puri, CEO, Ben Bolles, Executive Director, Product Management, and George Wagner, Director of Product and Technical Marketing at Liqid.


Liqid’s composable infrastructure hardware and software platform enable users to manage, scale-out, and configure physical bare-metal servers in seconds. This enables enterprises to increase infrastructure flexibility, agility, and efficiency from core to edge.


Data Center Transformation


Over time, data centers have transformed from a static, three-tier, and manually configured to hyper-converged and pre-configured, to what we have today -- software composable and dynamically configured.


The benefits of a composable infrastructure include: dynamically configured servers, independent scaling of resources, decoupling purchasing decisions, extending product lifecycles, enabling new configurations, improving software license efficiency, pay as you grow, and improving resource utilization 2X to 3X.


Brownfield applications have been a lucrative arena recently due to supply chain issues. At the end of the day, every company is interested in increasing the resource utilization of very expensive hardware.


Liqid promises cloud flexibility on-prem at a lower cost with prices running 69% to 72% lower than AWS for 4 GPU and 8 GPU, respectively.


Vision


Liqid's vision is to enable every server with composable software to address the low infrastructure utilization rate. $266.4 billion is spent on data centers; however, a lot of that money is wasted since 44% are non-production, 40% are oversized, 76% have idle time, and 50% are overspent. Liqid's goal is to help companies achieve up to 90% data center utilization.


In 2012, there were 500,000 data centers. Today there are 8 million+. Data centers are projected to use 20% of the world's electricity by 2025. Data centers use massive amounts of water - 1,600 trillion gallons per year (1MW = 8.5M gallons/year). Liqid enables up to 2 to 3X datacenter utilization and more than a 75% reduction in power.


Use Cases


Use cases where Liqid is getting the most traction include AI and deep learning, high-performance computing, dynamic cloud, 5G edge, and media and entertainment.

Composable architecture benefits high-value, mission-critical workloads.


Different workloads require different hardware profiles. Composable infrastructure handles the workload at hand and can then be repurposed for the next workload. This increases the ability to engage in more parallel workloads. Reconfigurability is dynamic requiring about 10 seconds without powering down the server.


For example, a composable AI platform needs to be able to handle data ingestion, data cleaning and tagging, training, and inference. Each of these four steps requires a different resource allocation. Liqid enables dynamic resource allocation for each stage of the AI workflow.


There is a huge opportunity for infrastructure on the edge with AI, GPUs, and recognition at the edge limited by power, cooling, and human access. The composable infrastructure enables you to take a box of GPUs, move them where they are needed, and increase utilization by 50 to 60%. This is a huge improvement in performance at the edge.


CDI Deployment over Fabric


The Liqid composable elements include CDI software for composition, fabric integrated PCIe expansion, and storage and memory.CDI deployment over fabric scales from 1 to 16 hosts, reconfigures in seconds, provides flexible CPU:GPU ratios, and enables users to take 20 GPUs and put them in a 1 GPU server.


Liqid enables the deployment of large-scale CDI environments with NVMe flash, storage-class memory, expansion, and hosts. It also enables high-performance scale-out solutions, impossible GPU configurations, and world-record AI benchmark performance.


Case Studies


Durham University has the COSmology MAchine (COSMA8), an HPC system for studying the origin and development of the universe. They needed to increase the speed and efficiency of COSMA8 by improving the utilization of GPUs. They chose Liqid to maximize the use of a small number of GPUs where they can populate servers with GPUs with the click of a button.

"Durham University is using cutting-edge CDI to accelerate research, improve resource utilization, and reduce the university's carbon footprint." -- Alastair Basden, Technical Manager, DiRAC Memory Intensive Service at Durham University

The U.S. Army Research Laboratory chose Liqid as their prime contractor to help them consume more than 100K CPU cores and approximately 300X GPUs. ARL is one of the top 200 supercomputers.


The U.S. Army Core of Engineers Engineer Research and Development Center also chose Liqid as their prime contractor for more than 85K CPU cores and approximately 600X GPUs.


Orange is the top telco provider in EMEA with more than 300,000 employees worldwide. Their goal is to deliver faster, smarter, and more efficient edge service. Their first goal

was to replace cloud-based AI workloads with on-prem deployments. Their second goal was to offer GPU as a service for edge deployments -- a premium product offering. They are using Liqid to deliver their dynamic edge pod which will enable the flexibility needed to meet the most demanding SLAs.