Result description
uHyper is a workload-aware, user-space hypervisor that eliminates performance interference when parallel applications share HPC nodes. Developed and validated in the EUPILOT project (TRL 4), the stack unifies: (i) the nOS-V threading system integrated with the LLVM OpenMP runtime for seameless execution of OpenMP applications; (ii) an optimised Task-Aware MPI (TAMPI) library for optimized intra-node communications; (iii) a task-aware glibc that re-implements POSIX pthreads on nOS-V for drop-in adoption on parallel applications using pthreads API; and (iv) a novel Linux syscall that accelerates context switches by an order of magnitude. By reconciling kernel and runtime decisions in user space, uHyper mitigates “OS noise”, prevents oversubscription, and delivers predictable, energy-efficient execution on x86 and Risc-V multi-cores. Benchmarks on representative HPC workloads show better time-to-solution without any application code changes.
Addressing target audiences and expressing needs
- Business partners – SMEs, Entrepreneurs, Large Corporations
- Technology Transfer Expertise
- Business Angels
National labs, super-computing centres, and applied-research institutes that can integrate and validate uHyper on production clusters.
HPC system integrators, RISC-V platform vendors, independent software vendors (ISVs) and end-user organisations that will package, deploy, or rely on the stack.
Deep-tech and HPC-focused venture funds that could finance industrialisation or a spin-off commercialising support, integration services, and enterprise licences.
- Other Actors who can help us fulfil our market potential
- Research and Technology Organisations
- Private Investors
R&D, Technology and Innovation aspects
All core components work together in lab-scale prototypes and have been experimentally validated on multi-core and RISC-V testbeds (TRL4). Next step, TRL 5-6 by hardening the codebase, completing a security review/certification, and staging pilots that scale from hundreds to thousands of cores. To fund this transition, and move on to TRL 7, we are exploring different funding opportunities.
Result submitted to Horizon Results Platform by BARCELONA SUPERCOMPUTING CENTER CENTRO NACIONAL DE SUPERCOMPUTACION
