Coursera Spring Sale
40% Off Coursera Plus Annual!
Grab it
Learn about GRANNY, a distributed runtime system that enables fine-granular management of compute-intensive applications in cloud environments through this 18-minute conference presentation from NSDI '25. Discover how traditional cloud schedulers fail to efficiently manage parallel applications implemented with multi-threading (OpenMP) or multi-processing (MPI), leading to scaling limitations and resource fragmentation. Explore GRANNY's innovative WebAssembly-based execution abstraction called Granules, which can execute application code with both thread and process semantics while enabling efficient snapshotting capabilities. Understand how GRANNY addresses key challenges by supporting vertical scaling of multi-threaded applications within VMs and horizontal migration of multi-process applications between VMs. Examine the system's ability to dynamically add Granules at runtime for scaling and migrate Granules between VMs to reduce fragmentation, launching new instances efficiently from snapshots. Review evaluation results demonstrating GRANNY's performance improvements, including up to 60% reduction in makespan for OpenMP workloads and up to 25% reduction in fragmentation for MPI workloads compared to current schedulers, presented by researchers from Imperial College London, INESC-ID, and Instituto Superior Técnico.