samba - Oct 2020 - Performance Question: Lots of Small Files vs One Large File

Is this a protocol issue? A decade ago I saw writes to small files less
than 16k were awful, because the cost of opening and other file ops dwarfed
writing actual content. So very small files should be slow, but these files
are 10mb each.

Or is this a Windows issue? If so, what's causing the problem?

Just trying to understand.

On Tue, Oct 6, 2020 at 2:39 PM Ralph Boehme <slow at samba.org> wrote:
> Am 10/6/20 um 7:49 PM schrieb Robert Buck via samba:
> > In an architecture where we have Samba running on RHEL exposing shares
to
> > Windows, when we have 1 large 1GB file, the write performance to
storage
> is
> > very fast, even over distances of 5000 miles. However, even writes to
> local
> > Samba servers, with 100 10MB files being copied onto a shared drive,
> > Windows Explorer is MUCH slower. I don't know if it's really
Samba, but
> > more than likely Windows.
> >
> > Does anyone on channel have experience with copying multiple small
files
> > onto shares? Are there any ways to make copying files onto a share
> faster?
>
> to save you a lot rocking back and forth, yes, small files workloads
> will be slower. The smaller, the slower... :)
>
> -slow
>
> --
> Ralph Boehme, Samba Team                https://samba.org/
> Samba Developer, SerNet GmbH   https://sernet.de/en/samba/
> GPG-Fingerprint   FAE2C6088A24252051C559E4AA1E9B7126399E46
>
> --
BOB BUCK
SENIOR PLATFORM SOFTWARE ENGINEER

SKIDMORE, OWINGS & MERRILL
7 WORLD TRADE CENTER
250 GREENWICH STREET
NEW YORK, NY 10007
T  (212) 298-9624
ROBERT.BUCK at SOM.COM

Am 10/6/20 um 9:11 PM schrieb Robert Buck:
> Is this a protocol issue? A decade ago I saw writes to small files less
> than 16k were awful, because the cost of opening and other file ops
> dwarfed writing actual content. So very small files should be slow, but
> these files are 10mb each.
> 
> Or is this a Windows issue? If so, what's causing the problem?
opening a file is expensive due to the WIndows semantics emulation layer
in Samba. It's also done sequentially in the server which means there's
not going to be much in-flight IO to keep the pipes full.

-slow

-- 
Ralph Boehme, Samba Team                https://samba.org/
Samba Developer, SerNet GmbH   https://sernet.de/en/samba/
GPG-Fingerprint   FAE2C6088A24252051C559E4AA1E9B7126399E46

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: OpenPGP digital signature
URL:
<http://lists.samba.org/pipermail/samba/attachments/20201006/3b45d1d5/signature.sig>

Am 06.10.20 um 21:11 schrieb Robert Buck via samba:
> Is this a protocol issue? 
> Or is this a Windows issue?
no, it's a general issue even without networking, no matter windows, 
linux or mac - a ton of small files are always a lot slower than a large 
chunk and it was always that way

it's unavoidable overhead
> On Tue, Oct 6, 2020 at 2:39 PM Ralph Boehme <slow at samba.org>
wrote:
> 
>> Am 10/6/20 um 7:49 PM schrieb Robert Buck via samba:
>>> In an architecture where we have Samba running on RHEL exposing
shares to
>>> Windows, when we have 1 large 1GB file, the write performance to
storage
>> is
>>> very fast, even over distances of 5000 miles. However, even writes
to
>> local
>>> Samba servers, with 100 10MB files being copied onto a shared
drive,
>>> Windows Explorer is MUCH slower. I don't know if it's
really Samba, but
>>> more than likely Windows.
>>>
>>> Does anyone on channel have experience with copying multiple small
files
>>> onto shares? Are there any ways to make copying files onto a share
>> faster?
>>
>> to save you a lot rocking back and forth, yes, small files workloads
>> will be slower. The smaller, the slower... :)

We're finding the issue is ctdb. We're looking at alternatives and
charges
in architecture. If your team has ideas in addition to what we're thinking
of please let us know. Thanks

On Tue, Oct 6, 2020 at 4:31 PM Reindl Harald <h.reindl at thelounge.net>
wrote:
>
>
> Am 06.10.20 um 21:11 schrieb Robert Buck via samba:
> > Is this a protocol issue?
> > Or is this a Windows issue?
>
> no, it's a general issue even without networking, no matter windows,
> linux or mac - a ton of small files are always a lot slower than a large
> chunk and it was always that way
>
> it's unavoidable overhead
>
> > On Tue, Oct 6, 2020 at 2:39 PM Ralph Boehme <slow at samba.org>
wrote:
> >
> >> Am 10/6/20 um 7:49 PM schrieb Robert Buck via samba:
> >>> In an architecture where we have Samba running on RHEL
exposing shares
> to
> >>> Windows, when we have 1 large 1GB file, the write performance
to
> storage
> >> is
> >>> very fast, even over distances of 5000 miles. However, even
writes to
> >> local
> >>> Samba servers, with 100 10MB files being copied onto a shared
drive,
> >>> Windows Explorer is MUCH slower. I don't know if it's
really Samba, but
> >>> more than likely Windows.
> >>>
> >>> Does anyone on channel have experience with copying multiple
small
> files
> >>> onto shares? Are there any ways to make copying files onto a
share
> >> faster?
> >>
> >> to save you a lot rocking back and forth, yes, small files
workloads
> >> will be slower. The smaller, the slower... :)
>
> --
BOB BUCK
SENIOR PLATFORM SOFTWARE ENGINEER

SKIDMORE, OWINGS & MERRILL
7 WORLD TRADE CENTER
250 GREENWICH STREET
NEW YORK, NY 10007
T  (212) 298-9624
ROBERT.BUCK at SOM.COM