Unix system call overhead

Question

About how many clock cycles does it take to make a simple, non-blocking system call on Linux?

Would I expect anything different from OS X?

I wonder about calls such as setsockopt, which might be used to provide an optional hint to improve performance, yet has its own intrinsic cost.

Please explain more what exact `setsockopt` call you want to make — Basile Starynkevitch, Mar 29 '14 at 16:03
@BasileStarynkevitch Say, `SO_SNDLOWAT`. It should just be setting an internal variable and returning. — Potatoswatter, Mar 30 '14 at 02:06
Related: [Fastest Linux system call](https://stackoverflow.com/q/48913091) (Meltdown / Spectre mitigations have made system calls more expensive.) — Peter Cordes, Apr 10 '22 at 10:27

score 6 · Accepted Answer · answered Mar 29 '14 at 16:03

6

It really depends on the system call and on the hardware.

The overhead of making a syscall is not very big! this gives less than a microsecond (i.e. a hundred processor clock cycles). Some syscalls are using vdso(7) to reduce that overhead.

However, for a given syscall, the kernel is doing some work. Depending on the actual involved syscall, this may take a lot of time.

For setsockopt(2) it should depend on which options you are setting.

In general, Linux socket implementation is quite good, and most of the time is spent on the actual network (i.e. Ethernet, etc....).

answered Mar 29 '14 at 16:03

Basile Starynkevitch

223,805
18
296
547

I'd actually expect Mac OS X finishing (respectable) second due to its hybrid kernel architecture. Of course it correct to assume that it really depends on the nature of the specific system call. – mfro Mar 29 '14 at 19:05
2

Update: Spectre / Meltdown mitigation make system calls significantly more expensive. Like [1800 cycles on a Skylake](https://stackoverflow.com/questions/48913091/fastest-linux-system-call#comment84843442_48914200), most of it in the `wrmsr` to flush branch predictors. – Peter Cordes Aug 05 '18 at 23:41

Unix system call overhead

1 Answers1