performance.md

General observability tools

• cat /proc/loadavg
• vmstat, dstat
• top, htop, atop
• collectl
• ps
• glances
• vtop
• systat (freebsd)

Tracing & profiling

• strace & truss & dtruss
• dtrace
• stap
• perf
• ltrace

#Cpu

• cat /proc/cpuinfo
• mpstat
• pidstat
• time (the binary is better than the shell built in)

#Memory

Gather info:

• cat /proc/meminfo
• pidstat gather info about memory usage of specific process for 60 second pertiods 5 times: pidstat -p <pid> -r 60 5

Tune swappiness /proc/sys/vm/swappiness show swap status swapon -s if OOM killer comes it kills based on OOM score, this can be seen also by top or by looking at /proc/<pid>/oom_score

#Storage

Gather info: /proc/diskstats

iostat
iotop
lsof
fuser
sar -d

#Network

All of these can be installed from a Debian package

• ss
• ifstat
• iptraf
• iftop
• vnstat
• bwm-ng
• speedometer

The universal tool - SAR

sar -n DEV

ss utility is a great replacement for netstat, it's faster and supports filtering. Make sure tcp_diag module is loaded otherwise it falls back to using /proc filesystem which makes it slower. Examples. get overall stats on network sockets

ss -s

Show all listening sockets

ss -l

Show only tcp listening sockets:

ss -tl

Filter by state

ss -ta state FIN-WAIT-1

#Gathering system information lshw or lshw -short gives a nice overview on the installed hardware dmidecode, dmidecode -t <n> where n is a number, see man page