Re: [PATCH 1/3] ws2_32: Implement a basic synchronous TransmitFile (try 3).

[Alexandre Julliard]

"Erich E. Hoover"  writes:
> - /* set reasonable defaults when requested */
> - if (!bytes_per_send)
> -
>    bytes_per_send = 1024;

Is that the Windows default?  It doesn't seem very reasonable to me, it
won't even fill a network packet.

[PATCH 1/3] ws2_32: Implement a basic synchronous TransmitFile (try 3).

[Erich E. Hoover]

On Mon, Oct 12, 2015 at 1:25 AM, Alexandre Julliard  wrote:
> "Erich E. Hoover"  writes:
> > - /* set reasonable defaults when requested */
> > - if (!bytes_per_send)
> > -
> >    bytes_per_send = 1024;
>
> Is that the Windows default?  It doesn't seem very reasonable to me, it
> won't even fill a network packet.

I don't know, it's not in the documentation.  However, if you like I
could try to tease it out by doing some tests and running Wireshark on
it.  I figured that since the default MTU is 1500 bytes that 1024
would be a reasonable power of two less than that to use, but if you
think that 2^16 is more reasonable then, obviously, we could use that
instead.

Best,
Erich

[PATCH 1/3] ws2_32: Implement a basic synchronous TransmitFile (try 3).

[Erich E. Hoover]

On Mon, Oct 12, 2015 at 9:16 AM, Erich E. Hoover
 wrote:
> On Mon, Oct 12, 2015 at 1:25 AM, Alexandre Julliard  wrote:
> > "Erich E. Hoover"  writes:
> > > - /* set reasonable defaults when requested */
> > > - if (!bytes_per_send)
> > > -
> > >    bytes_per_send = 1024;
> >
> > Is that the Windows default?  It doesn't seem very reasonable to me, it
> > won't even fill a network packet.
>
> I don't know, it's not in the documentation.  However, if you like I
> could try to tease it out by doing some tests and running Wireshark on
> it.  I figured that since the default MTU is 1500 bytes that 1024
> would be a reasonable power of two less than that to use, but if you
> think that 2^16 is more reasonable then, obviously, we could use that
> instead.

Some googling suggests that this value is controlled by
HKEY_LOCAL_MACHINE\System\CurrentControlSet\Services\AFD\Parameters\DefaultSendWindow
and that the default is different depending on the Windows version.
Apparently "normal" versions default to 64512 and server versions
default to 655360.  How would you like this to be approached?

Best,
Erich

[PATCH 1/3] ws2_32: Implement a basic synchronous TransmitFile (try 3).

[Bruno Jesus]

On Monday, October 12, 2015, Erich E. Hoover <
[e-mail]> wrote:
> On Mon, Oct 12, 2015 at 9:16 AM, Erich E. Hoover
> <[e-mail] javascript:;> wrote:
> > On Mon, Oct 12, 2015 at 1:25 AM, Alexandre Julliard <[e-mail]
>
> javascript:;> wrote:
> > > "Erich E. Hoover" <[e-mail] javascript:;>
>
> writes:
> > > > - /* set reasonable defaults when requested */
> > > > - if (!bytes_per_send)
> > > > -
> > > >    bytes_per_send = 1024;
> > >
> > > Is that the Windows default?  It doesn't seem very reasonable to me, it
> > > won't even fill a network packet.
> >
> > I don't know, it's not in the documentation.  However, if you like I
> > could try to tease it out by doing some tests and running Wireshark on
> > it.  I figured that since the default MTU is 1500 bytes that 1024
> > would be a reasonable power of two less than that to use, but if you
> > think that 2^16 is more reasonable then, obviously, we could use that
> > instead.
>
> Some googling suggests that this value is controlled by
>
> HKEY_LOCAL_MACHINE\System\CurrentControlSet\Services\AFD\Parameters\DefaultSendWindow
> and that the default is different depending on the Windows version.
> Apparently "normal" versions default to 64512 and server versions
> default to 655360.  How would you like this to be approached?

Hi all. What about reading SO_SNDBUF where it is available and 1024 where
it is not. Because I think send() will fail if you try to write more than
the kernel buffer size.

[PATCH 1/3] ws2_32: Implement a basic synchronous TransmitFile (try 3).

[Erich E. Hoover]

On Mon, Oct 12, 2015 at 8:03 PM, Bruno Jesus  wrote:
> ...
> Hi all. What about reading SO_SNDBUF where it is available and 1024 where it
> is not. Because I think send() will fail if you try to write more than the
> kernel buffer size.

I found some more useful information on what TransmitFile it supposed
to do (2^16 is a sane default):
http://source.winehq.org/patches/data/115187

We could consider using the SO_SNDBUF size, but how we treat it would
need to be platform dependent.  On Linux you need to divide the value
by 2 and on BSD it's the value it says it is - I'm not sure about
other platforms.  It might make sense to go to the trouble to do this,
but I'm not sure it's worth worrying about.  We could go back to 1024
(the minimum permitted value for the "real" SO_SNDBUF size), though,
as AJ points out, that value is quite small.  Since the "real" default
on Linux systems is 2^16 (2*2^16 reported) or larger I think we just
go with the 2^16 value from MSDN.

Best,
Erich