Hi,

I guess nobody ever checked them out.

For at least sendfile: I believe that the first kernels that came with IPFire 2.1 did not have an implementation of sendfile.

I additionally found, that the buffers are set to approx. 8kb which could also be increased. Supposedly, everyone has got enough of RAM.

Please open a bug to track this feature and prepare a patch.

Michael

On Thu, 2012-03-22 at 10:08 +0100, WhyTea wrote:

Hi again!

I've tested it again. Now with 1GBit/s

without these parameters the write speed is about 23 MByte/s withe these parameters the write speed is about 80 MByte/s

So I ask again, is there any reason why we don't use them?

• Daniel

-------- Original-Nachricht -------- Betreff: samba tuning Datum: Wed, 14 Mar 2012 13:34:17 +0100 Von: Daniel Weismüller whytea@ipfire.org Antwort an: whytea@ipfire.org An: development@lists.ipfire.org

Hi all,

is there any reason why ipfire don't use the following parameters with samba?

use sendfile = 1 aio read size = 1 aio write size = 1

I've tested them and it brings my samba-performance from 8,9 MB/s up to 9,2 MB/s (using 100Mbit/s fullduplex)

• Daniel

---

Development mailing list Development@lists.ipfire.org http://lists.ipfire.org/mailman/listinfo/development

On Thu, 2012-03-22 at 11:50 +0100, Christian Schmidt wrote:

Refering to http://www.samba.org/samba/docs/man/manpages-3/smb.conf.5.html

the better and correct syntax is

use sendfile = yes aio read size = 16384 aio write size = 16384

since sendfile can result in some problems with older clients I´am not quite sure about that

That's a good point. The documentation says there a issues with clients like Win 9x. Nobody does use that anymore. For those who still do, the option could be configurable. One still needs to decide what the default setting is. I would vote for "enabled".

The buffers and tcp options can be changed by using the socket option in the gui, the detault settings are taken quite conservative

True. Would this be a problem to change that? I don't think we should expect any regressions by that.

Michael