preadv2/pwritev2 rename

preadv2/pwritev2 rename

[Milosz Tanski]

Christoph and/or Jeff,

I updated the patch for 3.18-rc1 and I'm going to resend it as non-RFC
as I didn't get comments last time.

I only have one stupid question... I'm going to rename the calls to
preadv6 and pwritev6 (so it's more like the other syscalls: dup3,
accept4, eventfd2) but I'm not sure if i should call it preadv5 or
pwritev6 since the offset argument is split into two different
arguments (upper and lower part).

Also, In our application we were able to get about 20%-30% reduction
in response time when using this before queuing in a IO thread pool on
the read path. It's a pretty nice win in the real world.

Best,
- Milosz

-- 
Milosz Tanski
CTO
16 East 34th Street, 15th floor
New York, NY 10016

p: 646-253-9055
e: milosz-B5zB6C1i6pkAvxtiuMwx3w@public.gmane.org

Re: preadv2/pwritev2 rename

[Michael Kerrisk (man-pages)]

Hello Milosz,

On Mon, Oct 20, 2014 at 11:52 PM, Milosz Tanski <milosz-B5zB6C1i6pkAvxtiuMwx3w@public.gmane.org> wrote:
> Christoph and/or Jeff,
>
> I updated the patch for 3.18-rc1 and I'm going to resend it as non-RFC
> as I didn't get comments last time.
>
> I only have one stupid question... I'm going to rename the calls to
> preadv6 and pwritev6 (so it's more like the other syscalls: dup3,
> accept4, eventfd2) but I'm not sure if i should call it preadv5 or
> pwritev6 since the offset argument is split into two different
> arguments (upper and lower part).

It's points like this that show exactly why naming system calls after
the number of their arguments is a very bad idea[1]. Please don't do
it. pwritev2() and preadv2() are not pretty either, but are marginally
better. pwritev_fl() and preadv_fl() (or simialr) might also be okay,
I guess.

> Also, In our application we were able to get about 20%-30% reduction
> in response time when using this before queuing in a IO thread pool on
> the read path. It's a pretty nice win in the real world.

Cheers,

Michael

http://blog.man7.org/2014/02/system-call-naming-and-numbering.html

Re: preadv2/pwritev2 rename

[Michael Kerrisk (man-pages)]

On Tue, Oct 21, 2014 at 8:42 AM, Michael Kerrisk (man-pages)
<mtk.manpages-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> wrote:
> Hello Milosz,
>
> On Mon, Oct 20, 2014 at 11:52 PM, Milosz Tanski <milosz-B5zB6C1i6pkAvxtiuMwx3w@public.gmane.org> wrote:
>> Christoph and/or Jeff,
>>
>> I updated the patch for 3.18-rc1 and I'm going to resend it as non-RFC
>> as I didn't get comments last time.
>>
>> I only have one stupid question... I'm going to rename the calls to
>> preadv6 and pwritev6 (so it's more like the other syscalls: dup3,
>> accept4, eventfd2) but I'm not sure if i should call it preadv5 or
>> pwritev6 since the offset argument is split into two different
>> arguments (upper and lower part).
>
> It's points like this that show exactly why naming system calls after
> the number of their arguments is a very bad idea[1]. Please don't do
> it. pwritev2() and preadv2() are not pretty either, but are marginally
> better. pwritev_fl() and preadv_fl() (or simialr) might also be okay,
> I guess.
>
>> Also, In our application we were able to get about 20%-30% reduction
>> in response time when using this before queuing in a IO thread pool on
>> the read path. It's a pretty nice win in the real world.
>
> Cheers,
>
> Michael
>
> http://blog.man7.org/2014/02/system-call-naming-and-numbering.html

Also, please ensure that future iterations of these patches CC
linux-api@ as per Documentation/SubmitChecklist. The past ones did
not, and so this is the first mention of these system calls that I
happened to see.

Cheers,

Michael
-- 
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
Linux/UNIX System Programming Training: http://man7.org/training/

Re: preadv2/pwritev2 rename

[H. Peter Anvin]

On 10/20/2014 11:42 PM, Michael Kerrisk (man-pages) wrote:
> Hello Milosz,
> 
> On Mon, Oct 20, 2014 at 11:52 PM, Milosz Tanski <milosz-B5zB6C1i6pkAvxtiuMwx3w@public.gmane.org> wrote:
>> Christoph and/or Jeff,
>>
>> I updated the patch for 3.18-rc1 and I'm going to resend it as non-RFC
>> as I didn't get comments last time.
>>
>> I only have one stupid question... I'm going to rename the calls to
>> preadv6 and pwritev6 (so it's more like the other syscalls: dup3,
>> accept4, eventfd2) but I'm not sure if i should call it preadv5 or
>> pwritev6 since the offset argument is split into two different
>> arguments (upper and lower part).
> 
> It's points like this that show exactly why naming system calls after
> the number of their arguments is a very bad idea[1]. Please don't do
> it. pwritev2() and preadv2() are not pretty either, but are marginally
> better. pwritev_fl() and preadv_fl() (or simialr) might also be okay,
> I guess.
> 

The splitting of the argument is a calling convention thing (and a
rather stupid one at that... we shouldn't do these kinds of things
manually.)  As such, it is not visible to the user and should not be
counted.

	-hpa