STANNIX Solutions for the Real World


Manual: syscount.bt

syscount.bt(8)				   System Manager's Manual			      syscount.bt(8)

NAME
       syscount.bt - Count system calls. Uses bpftrace/eBPF.

SYNOPSIS
       syscount.bt -- [--sysname]

DESCRIPTION
       This  counts  system calls (syscalls), printing a summary of the top ten syscall IDs, and the top ten
       process names making syscalls. This can be helpful for characterizing the kernel and  resource  work‐
       load, and finding applications who are using syscalls inefficiently.

       This works by using the tracepoint:raw_syscalls:sys_enter tracepoint.

       Since this uses BPF, only the root user can use this tool.

OPTIONS
       --sysname Show the syscall name instead of syscall number.

REQUIREMENTS
       CONFIG_BPF and bpftrace.

EXAMPLES
       Count system calls until Ctrl-C is hit:
	      # syscount.bt

OUTPUT
       Top 10 syscalls IDs:
	      This  shows  the syscall ID number (in @syscall[]) followed by a count for this syscall during
	      tracing. syscount.bt defaults to outputting the syscall number.

       Top 10 syscalls NAMEs:
	      This shows the syscall ID name (in @syscallname[]) followed by a count for this syscall during
	      tracing. Use the --sysname parameter to output the system call name.

       Top 10 processes:
	      This shows the process name (in @process[]) followed by a count of syscalls during tracing.

OVERHEAD
       For most applications, the overhead should be manageable if they perform 1000's or even	10,000's  of
       syscalls	 per  second. For higher rates, the overhead may become considerable. For example, tracing a
       microbenchmark loop of 4 million calls to geteuid(), slows it down by 2.4x. However, this  represents
       tracing	a  workload  that  has	a  syscall rate of over 4 million syscalls per second per CPU, which
       should not be typical (in one large cloud production environment, rates of between 10k  and  50k	 are
       typical, where the application overhead is expected to be closer to 1%).

       For  comparison,	 strace(1) in its current ptrace-based implementation (which it has had for decades)
       runs the same geteuid() workload 102x slower (that's one hundred and two times slower).

SOURCE
       This is from bpftrace.

	      https://github.com/bpftrace/bpftrace

       Also look in the bpftrace distribution for a companion _examples.txt file containing  example  usage,
       output, and commentary for this tool.

       This is a bpftrace version of the bcc tool of the same name.  The bcc version provides different com‐
       mand line options, and translates the syscall IDs to their syscall names.

	      https://github.com/iovisor/bcc

OS
       Linux

STABILITY
       Unstable - in development.

AUTHOR
       Brendan Gregg

SEE ALSO
       strace(1)

USER COMMANDS					 2018-09-06				      syscount.bt(8)