CALL FOR PAPERS 22nd Workshop on Virtualization in High-Performance Cloud Computing (VHPC '27) Special focus 2027: virtualization for AI infrastructure -- memory, accelerators, state, and isolation held in conjunction with (Proceedings: ) ==================================================================== Date: Workshop URL: vhpc dot org Paper Submission Deadline: , 23:59 AoE Rolling abstract submission Abstract/Paper Submission Link: Call for Papers Containers and virtualization technologies constitute key enabling factors for flexible resource management in modern supercomputing centers and cloud environments alike. Operators must manage complex infrastructures of CPUs, GPUs and other accelerators, high-speed interconnects and tiered memory and storage, while supporting highly dynamic and heterogeneous workloads: tightly coupled simulations, data analytics pipelines, interactive services and, increasingly, large-scale AI training and inference. HPC centers are turning their systems into multi-tenant platforms and AI factories that serve many user communities at once, and cloud providers are building supercomputer-class AI clusters. In both worlds, virtualization and containerization decide who gets which resources, how well tenants are isolated from one another, and how computation and its state can be scheduled, checkpointed, moved and recovered. Various virtualization-containerization technologies contribute to the overall picture in different ways: machine virtualization, with its capability to consolidate heterogeneous software stacks, to isolate tenants with strong guarantees and to live-migrate a fully operating virtual machine (VM) with very short downtime, enables dynamic ways to manage physical servers; OS-level virtualization (i.e., containerization), with its capability to isolate multiple user-space environments within the same OS kernel, has become the standard way to deliver portable and reproducible software environments on HPC systems, provided it integrates with MPI, high-speed interconnects, GPU drivers and parallel filesystems. I/O virtualization allows network interfaces and accelerators to be shared by multiple VMs or containers; network virtualization creates logical overlays and tenant isolation independent of the underlying physical topology; and confidential computing extends isolation to protect data and models even from the infrastructure operator. Large-scale AI has made these questions central again, and the systems community is re-inventing classic virtualization ideas for accelerators: paging and tiering for the KV cache of large language models, oversubscribed, unified and disaggregated GPU memory, spatial and temporal multiplexing of GPUs among tenants and models, preemption and scheduling of accelerator work, and checkpoint, restore and migration of GPU state for fault tolerance, elasticity and fast start-up. At the same time, extensible kernels (eBPF, sched_ext) open the scheduler and the I/O path to new policies, and machine learning is itself being applied to managing virtualized infrastructure. VHPC '27 places a special focus on this intersection of virtualization and AI infrastructure, alongside its established scope of virtualization in HPC and cloud computing. Publication Accepted papers will be published in . Topics of Interest The VHPC program committee solicits original, high-quality submissions related to virtualization across the entire software stack, with a special focus on the intersection of HPC, containers- virtualization, the cloud and AI infrastructure. Major Topics: - HPC on containers and VMs - Virtualization for AI infrastructure (special focus 2027) - The supercomputer as a multi-tenant platform: HPC-as-a-Service, AI factories - Virtualization of accelerators, memory and interconnects each major topic encompassing design/architecture, management, performance measurement and modeling, and configuration/tooling: Special Focus 2027 -- Virtualization for AI Infrastructure: - Virtual memory for AI workloads: KV-cache paging, tiering and offloading; GPU unified memory and oversubscription - CXL-attached, pooled and disaggregated memory for training and inference - GPU and accelerator sharing and partitioning (MIG, MPS, vGPU, time-slicing); spatial and temporal multiplexing - GPU preemption, scheduling and admission control for multi-model, multi-tenant inference serving - Checkpoint, restore and live migration of GPU and accelerator state; fast model start-up and serverless GPUs - Fault tolerance, elasticity and malleability of large training jobs on virtualized infrastructure - Performance isolation and interference between tenants on shared AI systems - Confidential VMs and GPU trusted execution environments for AI, and their measured cost - Learned and LLM-assisted resource management, scheduling and configuration of virtualized infrastructure Design / Architecture: - Containers and OS-level virtualization for HPC (Apptainer, Podman, Sarus, Charliecloud, i.a.) - Containers at the system boundary: MPI and libfabric injection, ABI compatibility, GPU driver integration, parallel filesystems - Hypervisor and VMM design for HPC and AI (KVM, Xen, Cloud Hypervisor, i.a.) - Hypervisor support for heterogeneous resources (GPUs, NPUs, TPUs, FPGAs, DPUs/SmartNICs) - I/O virtualization: SR-IOV, Scalable IOV, vDPA, virtio, IOMMU - Network virtualization and tenant isolation on HPC interconnects (InfiniBand, Slingshot, RoCE) - Memory virtualization: nested paging, huge pages, unified CPU-GPU address spaces, CXL - VM and container trust and security models; confidential computing (AMD SEV-SNP, Intel TDX, ARM CCA) - Hypervisor and kernel mitigation of side-channel attacks - Extensible kernels: eBPF and sched_ext for scheduling, isolation and I/O - Cloud reliability, fault tolerance and high availability - Energy-efficient and power-aware virtualization - Virtualization on emerging architectures: ARM, RISC-V hypervisor extension, CPU-GPU superchips Management: - Container and VM management for HPC and cloud environments - Virtual clusters and tenant partitions on supercomputers; services next to batch (interactive, notebook, inference endpoints) - Batch systems and orchestrators together (Slurm, Flux, Kubernetes, Kueue, Dynamic Resource Allocation) - Topology-, NUMA- and accelerator-aware scheduling and placement - Co-scheduling of simulation, data analytics and AI workloads; preemption and malleable jobs - Workflow coupling with VMs and containers; in-situ analysis and AI coupled to simulation - Secure and trusted research environments for sensitive data on shared HPC systems - Novel models for non-HPC and AI workload provisioning on HPC resources - Policy, admission, quota and accounting for multi-tenant platforms - Power capping and per-tenant energy accounting Performance Measurements and Modeling: - Performance measurement, modeling and monitoring of virtualized and containerized HPC and AI workloads - Scalability analysis of VMs and containers at large scale - Virtualization overheads: TLB and page-walk costs, interrupt and timer virtualization, I/O paths - OS and virtualization noise, jitter and their mitigation - Interference and isolation between co-located tenants and models - Optimizations of virtual machine monitors and hypervisors for HPC and AI - Performance of confidential computing for HPC and AI - Benchmarks and reproducibility studies for virtualized environments Configuration / Tooling: - Operating MPI and collective communication libraries in containers and VMs - Container image building, distribution and lazy loading at scale; software supply-chain security for HPC images - Reproducible software environments; containers together with Spack and module systems - Job scheduling, control, policy and container placement in virtualized environments - Software-defined networks and network virtualization - GPU and accelerator virtualization operationalization - Observability, profiling and debugging tools for virtualized and containerized environments - Operations and experience reports from HPC centers and cloud providers The Workshop on Virtualization in High-Performance Cloud Computing (VHPC) aims to bring together researchers and industrial practitioners facing the challenges posed by virtualization in order to foster discussion, collaboration, mutual exchange of knowledge and experience, enabling research to ultimately provide novel solutions for virtualized computing systems of tomorrow. The workshop will be one day in length, composed of 20 min paper presentations, each followed by 10 min discussion sections, plus lightning talks that are limited to 5 minutes. Presentations may be accompanied by interactive demonstrations. Deep Dive Talks In addition to paper presentations, VHPC '27 hosts invited deep dive talks on the internals of virtualization: hardware virtualization extensions, hypervisors, kernels, interconnects and accelerators. Proposals for deep dive talks are welcome; please contact the chairs. Important Dates - Abstract/Paper submission deadline - Acceptance notification - Workshop Day - Camera-ready version due Chair Michael Alexander (chair), Austrian Academy of Sciences, Austria Anastassios Nanos (co-chair), Nubificus Ltd., UK Program committee Paper Submission-Publication Papers submitted to the workshop will be reviewed by at least two members of the program committee and external reviewers. Submissions should include abstract, keywords, the e-mail address of the corresponding author, and must not exceed 12 pages, including tables and figures at a main font size no smaller than 11 point. Short papers of up to 6 pages presenting work in progress or operations and experience reports are also welcome. Submission of a paper should be regarded as a commitment that, should the paper be accepted, at least one of the authors will register and attend the conference to present the work. Accepted papers will be published in . The format must be according to the style. Initial submissions are in PDF; authors of accepted papers will be requested to provide source files. Format Guidelines: Abstract, Paper Submission Link: Lightning Talks Lightning Talks are non-paper track, synoptical in nature and are strictly limited to 5 minutes. They can be used to gain early feedback on ongoing research, for demonstrations, to present research results, early research ideas, perspectives and positions of interest to the community. Submit abstract via the main submission link. General Information The workshop is one day in length and will be held in conjunction with . Please contact the chairs ahead of time about presenting remotely. Follow VHPC updates: https://x.com/VHPCworkshop and https://bsky.app/profile/vhpc.bsky.social