> 
> 
> > This is *not* a microoptimization. Its the sort of real optimization I
> > keep mentioning. Eliminating copies is significant in networking
> > stacks. It isn't the same thing as making getpid() run two million
> > times a second.
> 
> We do all sort of optimizations.  
> 
> Besides, we do not measure getpid (), we measure read()ing 1 byte from
> a null device.
> 
In the past, we have been through that, but considering the issue of general
syscall overhead the lmbench null syscall benchmark isn't that good because the
other side-effects of the benchmark.  That is NOT to say that the lat_syscall
benchmark is useless -- it just doesn't measure the simplest syscall which
of course measuring the simple syscall only would increase the SNR for
the "syscall overhead" measurment.

The lat_syscall benchmark measures the unfortunate overhead of the VFS
layering that *BSD has.

John