Re: [PATCH 3/3] perf record: mmap output file

[Ingo Molnar <mingo@kernel.org>]

* David Ahern <dsahern@gmail.com> wrote:

> On 10/15/13 10:06 AM, Ingo Molnar wrote:

> > splice() is very fast and should be able to process a lot of pages in 
> > one go, so the feedback loop should be pretty weak. mmap() triggers 
> > kernel code as well, every time we run out of the 64 MB window we got 
> > to remap it, right?
> 
> 
> Yes, 1 mmap, 1 munmap for every 64MB. Compare to the write() case which 
> calls write() for each mmap each time through the mmap_read loop.
> 
> I am conjecturing that splice would follow the write model in the sense 
> of a ring buffer has N bytes, call splice to copy the data from the ring 
> buffer to the file. So, splice saves on the memcpy, but not the 
> syscalls.

Well, splice can be 64 MB granular as well - but it depends on how 
buffered the pipes are (splice always uses pipes to shuffle around pages).

It's possible to have pipes with larger buffering - see F_SETPIPE_SZ. The 
current limit (for unprivileged users?) is:

  fs/pipe.c:unsigned int pipe_max_size = 1048576;

[But I haven't actually used this to increase splice() size yet.]

Thanks,

	Ingo