Skip to content

README.md

Latest commit

 

History

History
111 lines (76 loc) · 3.68 KB

README.md

File metadata and controls

111 lines (76 loc) · 3.68 KB

Large/Huge Pages scripts

Automation helpers for enabling Large/Huge pages on Windows and Debian. These are one-time ops/setup scripts, not runtime code.

scripts/
  windows/Enable-LargePages.ps1
  linux/enable_hugepages.sh

What these do

  • Windows (Enable-LargePages.ps1): Grants the SeLockMemoryPrivilege (“Lock pages in memory”) to a user so apps that request large pages can use them.
  • Debian (enable_hugepages.sh): Reserves HugeTLB 2MiB pages, mounts /mnt/hugepages, adds a hugepages group, and sets Transparent Huge Pages (THP) mode (default madvise). Persists settings.

⚠️ Caution: HugeTLB pages are non-swappable. Reserve only what your system can spare.

Prerequisites

  • Windows: Run PowerShell as Administrator.
  • Debian: Run with sudo/root; systemd present (for THP persistence).

Usage

Windows (PowerShell)

# In an elevated PowerShell in scripts/windows/
Set-ExecutionPolicy -Scope Process -ExecutionPolicy RemoteSigned
.\Enable-LargePages.ps1                 # current user
.\Enable-LargePages.ps1 -User 'MACHINE\miner'  # specific account

After running: sign out and back in (or restart the service account) so the new privilege appears in the logon token.

Verify:

whoami /priv | findstr /I SeLockMemoryPrivilege

Debian (Bash)

# In scripts/linux/
chmod +x enable_hugepages.sh
sudo ./enable_hugepages.sh                # defaults (~25% RAM, THP=madvise)
sudo ./enable_hugepages.sh -n 1024 -u myuser -t madvise

Verify:

grep -E 'HugePages_(Total|Free|Rsvd|Surp)' /proc/meminfo
mount | grep ' /mnt/hugepages '
cat /sys/kernel/mm/transparent_hugepage/enabled

Note: Log out/in so your user picks up the hugepages group membership (or use newgrp hugepages).

Rollback

Windows

  • Open Local Security Policy -> Local Policies -> User Rights Assignment -> Lock pages in memory -> remove the user -> gpupdate /force.
  • Sign out/in.

Debian

# Release HugeTLB pages now and on reboot
echo 0 | sudo tee /proc/sys/vm/nr_hugepages
sudo rm -f /etc/sysctl.d/99-hugepages.conf
sudo sysctl --system

# Unmount and remove fstab entry
sudo umount /mnt/hugepages || true
sudo sed -i '\|^[^#].\s/mnt/hugepages\s\+hugetlbfs|d' /etc/fstab
sudo rmdir /mnt/hugepages 2>/dev/null || true

# Disable THP systemd unit if created
sudo systemctl disable --now thp-mode.service 2>/dev/null || true
sudo rm -f /etc/systemd/system/thp-mode.service
sudo systemctl daemon-reload

Why Huge/Large Pages for RandomX (Monero) miners

RandomX intentionally hammers memory with pseudo-random accesses to a ~2 GiB dataset. This pattern punishes normal 4 KiB pages (lots of page-table walks and TLB misses). Enabling Huge/Large pages (2 MiB) helps by:

  • Cutting TLB misses: Fewer pages cover the same dataset -> fewer page-table lookups, less latency.
  • Smoother, higher hashrate: Typically a single-digit to low double-digit (%) gain depending on CPU, RAM speed/timings, and BIOS settings.
  • Lower CPU overhead & jitter: Fewer kernel crossings and page faults -> more stable per-thread performance.
  • Better efficiency: More hashes per watt thanks to reduced memory-management overhead.
  • Pinned memory (Windows “Lock pages”): Prevents paging of the RandomX dataset, avoiding stalls.

Notes:

  • Linux: explicit HugeTLB is most deterministic; THP=madvise generally plays nicer with other workloads.
  • Windows: granting SeLockMemoryPrivilege merely enables large-page allocations; the miner still chooses to use them.
  • Gains vary by hardware and BIOS (NUMA, memory channels, RAM frequency/latency).