Asgard is a hardware-partitioning technique for terabyte-scale bare-metal cloud servers using the Midgard address space. It improves granularity for core assignment to clients in bare-metal cloud and reduces unnecessary Last-Level Cache (LLC) communication and physical sharing, while maintaining high associativity.
An in-depth report of Asgard can be read in Report.pdf, and a poster can be found in Poster.pdf.
cpp_trace_analyzercontains a C++ cache simulator that reads the generated QEMU memory access traces and simulates different multi-level partitioned cache architectures, producing statistics that are used in the report.qemu-traceris a Rust QEMU plugin that generates memory access trace files to be analyzed by the C++ cache simulator.plottingcontains Python scripts for producing the different graphs exposed in the report and poster.