Intelligent, real-time migration of scientific computing applications on commercial cloud-based HPC platforms

Federal agencies spent $19.2B on the cloud in 2023, this represents approximately $6.2B more than the government spent in fiscal 2022 and translates into a compound annual growth rate (CAGR) of 47%. The increasing complexity and cost of cloud computing represent a significant challenge for national labs, academic institutions, and private industry that rely on high-performance computing (HPC) for large-scale scientific simulations and data analysis. Exostellar (formerly Exotanium) is developing intelligent automation and dynamic resource scaling in the GovCloud for both CPU and GPU-based computing. This particular project enables entering into the GPU-as-a-service sector, a $3B market in 2023 with a CAGR projected over the next 6 years of 40%. Exostellar addresses this by developing technologies for dynamic resource scaling and cost-efficient, real-time migration of HPC applications across commercial cloud platforms. Leveraging its innovative nested virtualization and X-Containers technology, Exostellar introduced two key products in Phase I: X-Infrastructure Optimizer (X-IO) and X-Workload Optimizer (X-WO), which enhanced the usability of Spot Instances significantly, with cost reduction of up to 90% compared to regular On-Demand instances. In Phase I, Exostellarĺs first product, X-IO, allowed companies to utilize Spot Markets as a low-cost alternative to regular VM instances, while circumventing the threats of price spikes and premature termination that currently severely limit their usability. The second product, X-WO, addressed idle resource waste in the cloud by packing idle containers onto a small number of VMs during the idle period, minimizing the number of active VMs and thus reducing the cost of keeping services online. Exostellar achieved successful migration of DOE HPC workloads within public cloud environments with substantial cost savings (70%) and no performance sacrifice without changing a line of code in the applications. Finally, we established the feasibility of supporting GPU-based workloads, the addition of which provides a full suite of capabilities for almost any HPC application. In Phase II, Exostellar expanded its work by integrating support for GovCloud and Azure GovCloud, enabling national labs like Idaho National Laboratory (INL) to run their confidential HPC workloads. X-IO and X-WO will offer seamless migration and dynamic scaling, aiming to broaden their application to a wider variety of DOE-related workloads. The third product, X-GPU Optimizer (X-GO), will revolutionize resource allocation by dynamically managing GPU resources, significantly minimizing instances of idle and over-provisioned GPUs. The goal is to create a versatile and efficient system that optimizes GPU usage while catering to the evolving needs of machine learning tasks.