Learn NetBSD - Performance & Optimization

#Introduction

In episode 18 we took NetBSD into the virtualization world. Now it's time to make sure the running system runs as fast and efficiently as possible. In this episode we'll dissect performance and optimization on NetBSD — tuning sysctl for the kernel, VM, and network, monitoring the system with systat and top, and comparing FFS with ZFS as a storage option.

#Basic Principle: Measure First, Tune Later

Measure the baseline with monitoring tools.
Identify the bottleneck (CPU, memory, disk, network).
Tune the relevant parameters, one at a time.
Measure again to confirm the improvement.

Performing a sysctl check

sysctl hw.ncpu
sysctl hw.physmem

Example output

hw.ncpu = 8
hw.physmem = 8471793664

#Monitoring with top and systat

#top

top shows processes and resource usage in real time:

Monitoring processes in real time

top

Example top output

load averages:  0.13,  0.08,  0.05
10 processes:  9 running,  1 idle
CPU states:  2.4% user,  0.0% nice,  1.8% system,  0.0% interrupt,  95.8% idle
Memory: 4096M Act, 340M Inact, 370M Wired, 30M Exec, 3056M Free

#systat

systat is NetBSD's deeper monitoring tool. Some of its views:

View	Function
`systat vmstat`	Memory and paging statistics
`systat netstat`	Network traffic
`systat iostat`	Disk I/O statistics
`systat swap`	Swap usage

Monitoring disk I/O in real time

systat iostat

Example systat iostat output

     disk0      disk1
 KB/t  tps   KB/t  tps
 32.5  12.3  12.1  3.4

#sysctl Tuning

#General Kernel Tuning

Basic parameters that are often adjusted:

Displaying several kernel parameters

sysctl kern.maxvnodes
sysctl kern.maxfiles

Example output

kern.maxvnodes = 52326
kern.maxfiles = 10480

Adjusting the file and vnode limits

sysctl -w kern.maxfiles=524288
sysctl -w kern.maxvnodes=262144

#Virtual Memory Tuning

NetBSD manages memory diligently. For read/write-heavy workloads, watch caching and paging:

Viewing the VM settings

sysctl vm.vm.vm_page_free_min
sysctl vm.vm.vm_page_free_target

Example VM output

vm.vm.vm_page_free_min = 512
vm.vm.vm_page_free_target = 2048

#Network Tuning

The most influential network parameter is the buffer — too small causes retransmits, too large eats memory:

Viewing the network buffers

sysctl net.inet.tcp.sendspace
sysctl net.inet.tcp.recvspace

Example output

net.inet.tcp.sendspace = 32768
net.inet.tcp.recvspace = 65536

For high-speed links (1 Gbps and up), raise sendspace:

Adjusting the TCP buffers

sysctl -w net.inet.tcp.sendspace=131072

The rule of thumb for TCP buffers: about Bandwidth x RTT (the bandwidth-delay product). To determine the RTT, use ping:

Measuring RTT to a server

ping -c 5 10.0.0.2

Example ping output (min/avg/max)

round-trip min/avg/max/stddev = 0.421/0.513/0.672/0.079 ms

Warning

Only raise network buffers if you truly need to — large buffers consume kernel memory per connection. With thousands of connections, a 128 KB buffer each means hundreds of MB of memory. Tuning must be proportional to the scale of the workload.

#Making Tuning Permanent

sysctl -w changes vanish on reboot. For permanence, write them in /etc/sysctl.conf — this file is read at boot:

Contents of /etc/sysctl.conf

kern.maxfiles=524288
kern.maxvnodes=262144
net.inet.tcp.sendspace=131072

#Storage Tuning

#Filesystem Options

FFS accepts mount options that affect performance. Viewing the active options:

Viewing the active mount options

mount | grep ffs

Example output

/dev/wd0e on /usr type ffs (local, nodevmtime, noatime)

Options commonly considered:

Option	Effect
`noatime`	Doesn't update access times — reduces write I/O
`nodevmtime`	Doesn't write times to the device — similar to noatime
`norelatime`	Disables relatime (the default)

Example of adding noatime in /etc/fstab for a frequently-read /var partition:

fstab line with noatime

/dev/wd0f   /var         ffs     rw,noatime  1 2

#FFS vs ZFS

ZFS is available on NetBSD as a package (zfs). Both have different strengths:

Aspect	FFS	ZFS
Maturity on NetBSD	Default, very mature	Port, stable for general use
Data checksums	Not automatic	Yes (corruption detection)
Snapshots	Via fsck -t	Built-in, very easy
RAID/volume	Needs cciss/cgd + raframe	Built-in (mirror, raidz)
Complexity	Simple	More complex

Installing ZFS from pkgsrc

cd /usr/pkgsrc/filesystems/zfs
make install clean

#A Recommended Tuning Flow

Step	Action
1	Measure the baseline with `top`, `systat iostat`, `systat netstat`
2	Identify the dominant bottleneck
3	Change one parameter in `/etc/sysctl.conf`, then reboot or `sysctl -w`
4	Measure again and compare
5	Keep only the changes proven to help

#Closing

In this episode 19, you've squeezed performance out of NetBSD: monitoring the system with top and systat, tuning sysctl for the kernel, VM, and network, making tuning permanent in /etc/sysctl.conf, and comparing FFS with ZFS for storage needs.

Key takeaways:

Measure before tuning: baseline first, one change at a time, measure again.
Monitor with top (processes) and systat iostat/netstat (I/O, network).
Common tuning: kern.maxfiles, net.inet.tcp.sendspace — match buffers to the bandwidth-delay product.
Permanent tuning goes in /etc/sysctl.conf.
FFS for the default, ZFS when you need checksums/snapshots/integrated RAID.

In the next episode, episode 20, we'll dive into the soul of NetBSD: embedded, ARM, and a portability deep-dive — exploring architecture support like arm64, riscv, and evbarm, cross-arch building, and use on SBCs and legacy hardware. See you in episode 20!