PMCTrack v3.0 released

Published: Jan 4, 2023 by Juan Carlos Saez

A new version of PMCTrack has been released (v3.0). The new version features:

  • The PMCSched subsystem, an open-source framework enabling rapid-prototyping of OS-level scheduling-related support in Linux without patching the kernel. Many thanks to Carlos Bilbao (aka @Zildj1an) for creating this new exciting and powerful subsystem of PMCTrack !

    PMCSched currently supports systems with Intel, AMD or ARM (armv7) processors. Among other things, the framework allows:

    • The creation of OS scheduling algorithms as separate scheduling plugins (implemented inside PMCTrack’s kernel module). Because all these algorithms operate at kernel space, they could be leveraged in different scheduling scenarios, including workloads running natively on the system or inside containers, as well as handling multiple KVM-based virtual machines running on a single node
    • The implementation of custom interaction mechanisms between the OS and a user-level runtime system for cooperative improved scheduling on symmetric and asymmetric multicore systems
    • Experimenting with new hardware-aided technologies to assist a scheduler, such as Intel Thread Director.
    • The rapid-prototyping of OS-level resource-management schemes, also as separate plugins within PMCTrack’s kernel module. The implementation includes some sample LLC-partitioning algorithms (such as UCP and LFOC+).

  • PMCTrack support for newer kernels (latest version tested Linux v5.16.y)

  • Other new relevant features include:

    • Support for Intel Alder Lake processors and AMD EPYC Rome processors
    • A set of proof-of-concept PMCSched scheduling plugins
    • Support for the Intel Enhanced Hardware Feedback Interface (aka Intel Thread Director)
    • Support for latest Odroid kernel (v5.4.y)

Share