IntervalVM SOSP2025 Artifact

This is the artifact for the paper “Scalable Address Spaces using Concurrent Interval Skiplist”, accepted to the 31st Symposium on Operating Systems Principles (SOSP '25). It includes our modified Linux kernel as well as the benchmarks used in the paper.

For the complete Linux kernel tree, please refer to this repository.

Installation

Run

cd scripts && sudo ./install-all.sh

to install all required components (requires ~65 GiB of disk space). On our 48-core machine, this process takes about 1.5 hours and performs the following steps:

• Runs install-packages.sh to install required packages.

Note: this script uses apt-get and is compatible only with Ubuntu/Debian. For other distributions, edit install-packages.sh to use the appropriate package manager.

• Runs install-kernels.sh to install the Linux kernels used in our evaluation. This script sequentially executes install-linux-6.8.0.sh, install-linux-6.8.0-debug.sh, install-linux-ivm.sh, and install-linux-ivm-debug.sh.
• Runs install-benchmarks.sh to install the benchmarks. This script sequentially executes install-apache.sh, install-ds_benchmark.sh, install-leveldb.sh, install-lmbench.sh, install-metis.sh, install-microbench.sh, install-psearchy.sh, and install-parsec-benchmark.sh.

Alternatively, you may also install components individually by running the corresponding scripts.

Booting

To reproduce the figures in the paper, first boot into one of the newly installed kernels.

The following kernels are installed by install-kernels.sh:

• Linux 6.8.0 — baseline
• Linux 6.8.0-debug — baseline with debug tools enabled
• Linux 6.8.0-interval-vm+ — our customized kernel
• Linux 6.8.0-interval-vm-debug+ — our customized kernel with debug tools enabled

On Ubuntu, simply run one of the set-boot-kernel-as-*.sh in scripts to set the default boot kernel, and then execute sudo reboot. These scripts update GRUB_DEFAULT in /etc/default/grub and run sudo update-grub.

For other distributions, edit /etc/default/grub manually and add:

GRUB_TIMEOUT_STYLE=menu
GRUB_TIMEOUT=10   # Gives 10 seconds to choose

This enables selecting the desired kernel at boot.

Running Benchmarks

After booting into one of the new kernels, run:

./run-bench.sh

This script automatically detects the current kernel and executes the required benchmarks to produce results for the figures. See run-bench.sh for details.

Approximate runtime on our machine:

• Linux 6.8.0: 30 hours
• Linux 6.8.0-interval-vm+: 24 hours
• Linux 6.8.0-debug: 19 hours
• Linux 6.8.0-interval-vm-debug+: 9 hours

Alternatively, you can run individual benchmarks by navigating to one of the benchmark directories (apache, ds_benchmark, leveldb, lmbench, lockstat, metis, microbench, psearchy, parsec-benchmark) and executing:

sudo python3 bench.py

Exception: When booted with Linux 6.8.0-interval-vm-debug+, use:

sudo ./bench_debug.sh

instead of sudo python3 bench.py. Note that this script is available only in the apache and microbench directories.

Plotting the Results

After benchmarks complete, run:

python3 plot.py

in each benchmark directory to generate the plots as PDF files.

Exception: lmbench does not provide a plot.py. Instead, run:

python3 compare.py

to view results.

Benchmark Descriptions

For each benchmark, we provide:

• a summary of its purpose,
• the kernels under which it should be run,
• the command to execute it,
• the runtime on our machine, and
• instructions for comparing the results with the figures in the paper.

All benchmark results are stored in the results directory within each benchmark's directory.

apache

• Evaluates Apache web server using wrk with two configurations: 1) single process, 2) default.
• Run under: Linux 6.8.0, Linux 6.8.0-interval-vm+, or Linux 6.8.0-interval-vm-debug+
• Command:
  • sudo python3 bench.py for Linux 6.8.0 and Linux 6.8.0-interval-vm+
  • sudo ./bench_debug.sh for Linux 6.8.0-interval-vm-debug+
• Runtime: 2.5 hours for bench.py, 7 hours for bench_debug.sh
• Compare: apache.pdf to Fig. 15a, and results/default_results.csv to Fig. 16.

ds_benchmark

• Evaluates data structures: maple tree (mp) and interval skiplist (isl) across operations (Query, Alloc, Map).
• Run under: Linux 6.8.0
• Command: sudo python3 bench.py
• Runtime: 1 hour
• Compare latency.pdf to Fig. 12a, and Query.pdf, Alloc.pdf, Map.pdf to Fig. 12b.

leveldb

• Evaluates LevelDB using db_bench with the readrandom benchmark.
• Run under: Linux 6.8.0 or Linux 6.8.0-interval-vm+
• Command: sudo python3 bench.py
• Runtime: 6.5 hours for Linux 6.8.0, 1 hour for Linux 6.8.0-interval-vm+
• Compare leveldb.pdf to Fig. 15b.

lmbench

• Runs the lmbench suite multiple times.
• Run under: Linux 6.8.0 or Linux 6.8.0-interval-vm+
• Command: sudo python3 bench.py
• Runtime: 6 hours
• Compare the output of python3 compare.py to Fig. 13.

lockstat

• Runs apache, metis, psearchy, and leveldb while using lock_stat to measure wait time / total time.
• Run under: Linux 6.8.0-debug
• Command: sudo python3 bench.py
• Runtime: 24 hours
• Compare wasted_time_graph.pdf to Fig. 1.

metis

• Evaluates Metis Map-Reduce with the wrmem benchmark (4GiB input).
• Run under: Linux 6.8.0 or Linux 6.8.0-interval-vm+
• Command: sudo python3 bench.py
• Runtime: 3.5 hours
• Compare metis.pdf to Fig. 15c.

microbench

• Evaluates address space operations with microbenchmarks.
• Run under: Linux 6.8.0, Linux 6.8.0-interval-vm+, or Linux 6.8.0-interval-vm-debug+
• Command:
  • sudo python3 bench.py for Linux 6.8.0 and Linux 6.8.0-interval-vm+
  • sudo ./bench_debug.sh for Linux 6.8.0-interval-vm-debug+
• Runtime: 1.5 hours for bench.py, 2 hours for bench_debug.sh
• Compare Alloc.pdf and Alloc + Fault + Modify.pdf to Fig. 14.

psearchy

• Evaluates text indexing using Psearchy on the Linux 6.8.0 source tree.
• Run under: Linux 6.8.0 or Linux 6.8.0-interval-vm+
• Command: sudo python3 bench.py
• Runtime: 5 hours
• Compare psearchy.pdf to Fig. 15d.

parsec-benchmark

• Evaluates the overall performance of various multithreaded applications.
• Run under: Linux 6.8.0 or Linux 6.8.0-interval-vm+
• Command: sudo python3 bench.py
• Runtime: 4 hours
• Compare parsec.pdf to Fig. 16.

Uninstall

Run:

cd scripts && sudo ./uninstall-all.sh

Miscellaneous

Our artifact was evaluated in the following environment:

• Motherboard: Supermicro X11DPG-OT-CPU
• CPU: Dual-socket Intel Xeon Gold 6248R @ 3.0 GHz (48 cores)
• Memory: 384 GiB
• OS: Ubuntu 24.04.1 LTS

License

This artifact is distributed under GPL-2.0, except where otherwise noted. Some benchmarks are derived from prior benchmarks, which retain their original licenses. Where applicable, those licenses are included in the artifact.