CPU settings for production

Problem

CPU power-saving modes (such as Intel SpeedStep, AMD Cool'n'Quiet, C-states, and P-states) can hurt YDB performance for several reasons:

  1. Unstable CPU frequency — YDB expects stable CPU throughput for predictable operation.
  2. Core affinity issues — changing per-core frequency can break load balancing across threads.
  3. Mode-switching latency — transitions between power-saving states add extra delays.
  4. Unpredictable throughput — operations may run at different speeds depending on the current CPU mode.

Typical symptoms:

  • Unstable query latency.
  • Unexplained performance drops.
  • Uneven load across cluster nodes.
  • Erratic behavior under heavy load.

Solution

Recommended BIOS/UEFI settings

For YDB servers, configure the following BIOS/UEFI settings:

  1. CPU Power Management — set to Performance or Max Performance.
  2. Intel Turbo Boost — enable for maximum performance.
  3. C-states — disable or limit to C1/C1E.
  4. P-states — use a fixed maximum frequency.
  5. Intel SpeedStep / AMD Cool'n'Quiet — disable.
  6. Package C-State Limit — set to C0 or C1.

Operating system settings

On Linux, for deployments not using the Ansible role (which configures this automatically), run the following commands as root:

# Force the performance cpufreq governor
echo 'performance' | sudo tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor >/dev/null

# Pin minimum frequency to maximum
echo $(cat /sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_max_freq) | \
sudo tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_min_freq >/dev/null

# Reduce PCIe ASPM power saving
echo 'performance' | sudo tee /sys/module/pcie_aspm/parameters/policy >/dev/null

Verifying current settings

Useful checks (after applying settings, the first command should output performance for every CPU core):

# Current governor
cat /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor

# Available frequencies
cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies

# C-states
cat /sys/devices/system/cpu/cpu0/cpuidle/state*/name

Rationale

As a high-performance distributed database, YDB needs stable and predictable CPU performance. Client-style power-saving modes are a poor fit for server workloads that require sustained availability and low latency.

A stable CPU frequency helps with:

  • Predictable operation latency.
  • Correct behavior of load-balancing mechanisms.
  • Steady throughput under load.
  • Minimal delays for transaction processing.

Exceptions

Compromises may be acceptable when:

  • Test environments — energy efficiency matters more than raw performance.
  • Standby nodes — not serving active traffic.
  • Low-load deployments — power savings are a priority.

For busy production clusters, follow the recommendations above.

Previous
System requirements
Next
Versioning