Learning Devuan GNU/Linux - Performance & Tuning

#Introduction

A healthy system isn't just one that runs — it's one that runs efficiently. Episode 20 covers performance and tuning on Devuan: tuning the kernel with sysctl, choosing the right I/O scheduler, speeding up boot with a minimal init, and monitoring system health with htop, sysstat, psacct, and dmesg.

The good news: non-systemd systems have a hidden advantage — a simple boot is usually faster and easier to measure. You can see every boot step without systemd units hiding the process.

#Kernel Tuning with sysctl

#Reading and Setting Parameters

sysctl manages kernel parameters at runtime. Start by checking current values:

sysctl vm.swappiness
sysctl net.ipv4.tcp_congestion_control

sysctl vm.swappiness shows the kernel's tendency to use swap — Debian's default is 60. For servers with lots of RAM, a lower value (for example 10) reduces unnecessary swap writes.

#Applying Permanent Changes

Change a value temporarily, then make it permanent through a configuration file:

sudo sysctl -w vm.swappiness=10
echo "vm.swappiness=10" | sudo tee /etc/sysctl.d/99-tuning.conf
sudo sysctl --system

sudo sysctl -w vm.swappiness=10 applies it immediately; the second line saves it. sysctl --system reloads all configuration files — a way to apply changes without rebooting.

#Other Useful Parameters

Some parameters commonly tuned on Devuan servers:

vm.dirty_ratio = 20
vm.dirty_background_ratio = 5
net.core.somaxconn = 1024
fs.file-max = 65535

The vm.dirty_ratio and vm.dirty_background_ratio parameters control when the kernel writes dirty data to disk, while net.core.somaxconn and fs.file-max raise the connection and file limits for high workloads.

#I/O Scheduler

#Choosing the Right Scheduler

The I/O scheduler determines the order of disk reads and writes. Modern kernels use a per-device scheduler:

cat /sys/block/sda/queue/scheduler
echo none | sudo tee /sys/block/sda/queue/scheduler

cat /sys/block/sda/queue/scheduler shows the available schedulers with the active one in square brackets. For SSDs, none (or mq-deadline on older kernels) is usually best; for classic HDDs, bfq or mq-deadline is more responsive.

#Making the Choice Permanent

To make the choice survive a reboot, set it via a GRUB kernel parameter:

sudo sed -i 's/^GRUB_CMDLINE_LINUX=.*/GRUB_CMDLINE_LINUX="elevator=none"/' /etc/default/grub
sudo update-grub

update-grub rebuilds the GRUB menu. After reboot, the kernel uses the elevator=none parameter for all disks.

#Speeding Up Boot with a Minimal Init

#Auditing Boot Services

The fewer services enabled, the faster the boot. Audit the services that start at runlevel 2:

ls /etc/rc2.d/
service --status-all | grep "\[ + \]"

ls /etc/rc2.d/ lists the active service symlinks. Disable services you don't need:

sudo update-rc.d bluetooth disable
sudo update-rc.d cups disable

update-rc.d bluetooth disable prevents bluetooth from starting at boot. On a server, remove graphical and audio services from the default runlevel for the fastest boot.

#Measuring Boot Time

Measure boot duration to compare before and after tuning:

systemd-analyze 2>/dev/null || time (echo "boot selesai")
date -d @$(awk '{print $1}' /proc/uptime) -u +%T

/proc/uptime shows the time since boot in seconds — a simple way to measure boot duration. Record the value before and after changes to see the effect of tuning.

#Monitoring with sysstat and Other Tools

#Installing Monitoring Tools

sysstat provides sar for performance history; psacct records per-process resource usage:

sudo apt install sysstat htop psacct btop
sudo service sysstat start

sudo service sysstat start enables periodic sar data collection. Also enable psacct so user activity is recorded.

#Reading Performance

Monitor the system in real time and historically:

htop
sar -u 1 5
dmesg | grep -iE "error|warn" | tail

htop shows processes and CPU/RAM usage interactively. sar -u 1 5 takes five samples of CPU usage per second. For hardware diagnostics, dmesg | grep -iE "error|warn" shows the latest problems reported by the kernel.

#Conclusion

Episode 20 covered performance and tuning on Devuan: tuning kernel parameters with sysctl, choosing a per-device I/O scheduler, speeding up boot with a minimal init, and monitoring the system with htop, sysstat, psacct, and dmesg.

Key takeaways:

• sysctl tunes kernel parameters; save them in /etc/sysctl.d/.
• A swappiness of 10 reduces unnecessary swap writes.
• SSDs suit the none scheduler; HDDs suit bfq.
• Audit and disable services you don't need at boot.
• sysstat provides historical data through sar.
• dmesg is the first source for hardware diagnostics.

In the next episode, episode 21, we'll cover roadmap and community — the direction of Excalibur point releases and the testing suite, the role of ceres, installer and init tooling development, and Devuan community resources like the forum, mailing lists, and IRC.