README.md

HyperAlloc Benchmarks

This repository contains the benchmarks and results for the HyperAlloc memory reclamation approach.

Publication

HyperAlloc: Efficient VM Memory De/Inflation via Hypervisor-Shared Page-Frame Allocators Lars Wrenger, Kenny Albes, Marco Wurps, Christian Dietrich, Daniel Lohmann In: Proceedings of the 20th European Conference on Computer Systems (EuroSys 2025); ACM

Artifact Evaluation

The artifact evaluation instructions are in artifact-eval/.

Related Repositories

• hyperalloc-linux: Modified Linux kernel
• hyperalloc-qemu: Modified QEMU monitor
• linux-alloc-bench: Kernel module for benchmarking the page allocator
• hyperalloc-stream: STREAM memory bandwidth benchmark
• hyperalloc-ftq: FTQ CPU work benchmark
• llfree-c: C-based implementation of the LLFree page allocator
• llfree-rs: Rust-based implementation of the LLFree page allocator, including some micro benchmarks, like write.rs

Benchmark Data & Visualization

The benchmarks have been developed and tested on Linux 6.1 (Debian 12) for host and guest, and QEMU 8.2.50. Even though they should be mostly compatible with other Linux versions, VFIO might work slightly different.

Setup venv and install dependencies:

python3 -m venv venv
source ./venv/bin/activate
pip3 install -r requirements.txt

Run a benchmark:

sudo -E ./max_power.sh python compiling/compiling.py --mode <mode> --img <path/to/disk.qcow2> -c 8 -m 8

The max_power.sh script disables powersaving and frequency scaling on Intel CPUs.

The mode specifies the paths to the guest kernel and QEMU to be used. You can manually overwrite them with the --kernel and --qemu arguments. The benchmark mode can be one of the following:

• base-manual: Unmodified QEMU and guest with manual virtio-balloon
• base-auto: Unmodified QEMU and guest with enabled free-page-reporting
• huge-manual: Modified QEMU and guest which use huge pages instead of base pages for ballooning
• virtio-mem: Unmodified QEMU and guest with enabled virtio-mem, that adds movable memory
• llfree-manual: Modified QEMU and guest with integrated LLFree and manual reclamation.
• llfree-auto: Modified QEMU and guest with integrated LLFree and auto reclamation.

The disk image is expected to contain a debian 12 with a password-less "debian" user. Also the compile and multivm benchmarks expect a checked out clang 16.0.0 repository at ~/clang. The stream and ftq benchmarks require the STREAM and ftq repos at ~/STREAM and ~/ftq.

VFIO

Some benchmarks require VFIO devices. We would recommend using networks cards. For binding device groups to VFIO use bind_vfio.py, which lists all available groups and allows you to choose one.

sudo python3 bind_vfio.py

This only works if your system supports IOMMUs and for Intel, the IOMMUs have to be activated via the kernel commandline intel_iommu=on.

Then the devices from that vfio group can be given to the benchmark, which then passes it through to the VMs.

sudo -E ./max_power.sh python compiling.py ... --vfio-dev <device-id>
# or alternative bind all device from that group
sudo -E ./max_power.sh python compiling.py ... --vfio <group-number>