I/O pattern of smbd / optimization of disk subsystem

[Jones]

Hello MinWai,

Hmm i think so because after followed the instructions as the pdf said,
command filefrag told that my linux box got probably the same result.

Upload 1000 files with size 2.6MB (2682880 Bytes) as test patterns,
strict allocate = yes could ease the fragments.

And im not sure if this helps performance improvement,
it might take a bit to do some testing later.

my linux box:
file system: ext4
kernel-3.4.6
samba-4.0.5

--
Regards,
Jones

2014-10-05 22:30 GMT+08:00 Min Wai Chan <dcmwai at gmail.com>:

> Hi Jones,
>
> This is very interesting.
>
> Does that still apply today on the current version?
>
> Thank You.
>
> On Sun, Oct 5, 2014 at 9:29 PM, Jones <jones.kstw at gmail.com> wrote:
>
>> Hello Heiner,
>>
>> This link might be interesting:
>> http://sambaxp.org/fileadmin/user_upload/SambaXP2010-DATA/Olaf_Flebbe.pdf
>>
>> --
>> Regards,
>> Jones
>>
>> 2014-10-03 22:49 GMT+08:00 Heiner Lesaar <heiner.lesaar at googlemail.com>:
>>
>> >
>> > Dear list,
>> >
>> > I am keen to learn more about the characteristics on how smbd accesses
>> the
>> > disks in terms of request sizes. Does this depend on some of the
>> smb.conf
>> > parameters (max xmit etc.) or on the negotiated protocol with the smb
>> > client?
>> >
>> > Can someone give me a hint on how to find out what the characteristic in
>> > my usecase would be? I read about blktrace and collectl as some of the
>> > suggested tools. Usage is mainly multiple, slow but big streaming reads
>> and
>> > writes.
>> >
>> > My goal is to optimize the disk subsystem config (multiple raid5 sets)
>> to
>> > get the best performance from the disks without suffering from
>> side-effects
>> > like read/modify/write penalties when the data isn't written in matching
>> > allocation sizes to hit the "full stripe" of the disks. Same for read
>> > processes obviously.
>> >
>> >
>> > Thanks in advance for any input on this!
>> >
>> > Heiner
>> >
>> >
>>
>
>