samba - Aug 2023 - Increase data length for SMB2 write and read requests for Windows 10 clients

Hello Jeremy,
> OH - that's *really* interesting ! I wonder how it is
> changing the SMB3+ redirector to do this ?
It looks like applications could do something and give a hint to SMB3+ 
redirector, so far not quite sure how to make it,
per process monitor (procmon) could show that write I/O size seems 
could be pass from the application layers,
(https://learn.microsoft.com/en-us/sysinternals/downloads/procmon)
this is a example for writing, using applications (Fastcopy.exe, 
explorer.exe, and robocopy.exe) to transfer a file named '1GB.img' from 
local disk 'H:' to remote SMB mount disk 'Z', 
and monitor them with procmon: 

1. FastCopy (FastCopy.exe), 8MB (need to revise FastCopy default setup)
Process Name	Operation	Path	Detail
System	TCP Send	10.19.7.63:64845 -> 10.19.250.98:445	Length: 8388724,
startime: 3459115, endtime: 3459149, seqnum: 0, connid: 0
System	TCP Receive	10.19.7.63:64845 -> 10.19.250.98:445	Length: 84, seqnum:
0, connid: 0
FastCopy.exe	WriteFile	\\10.19.250.98\Public\1GB.img	Offset: 92,274,688, Length:
8,388,608, Priority: Normal

2. File Explorer (explorer.exe), 1MB, drag-and-drop files.
Process Name	Operation	Path	Detail
explorer.exe	ReadFile	H:\1GB.img	Offset: 55,574,528, Length: 1,048,576,
Priority: Normal
System	ReadFile	H:\1GB.img	Offset: 55,574,528, Length: 1,048,576, I/O Flags:
Non-cached, Paging I/O, Priority: Normal
System	WriteFile	\\10.19.250.98\Public\1GB.img	Offset: 55,574,528, Length:
1,048,576, Priority: Normal
System	TCP Send	10.19.7.63:64845 -> 10.19.250.98:445	Length: 1048692,
startime: 3545682, endtime: 3545689, seqnum: 0, connid: 0
System	TCP Receive	10.19.7.63:64845 -> 10.19.250.98:445	Length: 84, seqnum:
0, connid: 0

3. robocopy (robocopy.exe), 1MB, command 'robocopy.exe h: z: 1GB.img
/NP'
Process Name	Operation	Path	Detail
Robocopy.exe	ReadFile	H:\1GB.img	Offset: 520,093,696, Length: 1,048,576,
Priority: Normal
System	WriteFile	\\10.19.250.98\Public\1GB.img	Offset: 520,093,696, Length:
1,048,576, Priority: Normal
System	TCP Send	10.19.7.63:64845 -> 10.19.250.98:445	Length: 1048692,
startime: 3618938, endtime: 3618945, seqnum: 0, connid: 0
System	TCP Receive	10.19.7.63:64845 -> 10.19.250.98:445	Length: 84, seqnum:
0, connid: 0

Perhaps i could do 'procmon' with other benchmark tool like iometer,
get back here later :)

Found a video talking about SMB2 redirector and large I/O size,
thouhd it is not up-to-date SMB3+ redirector still good to know about 
related design thoughts in the history:
https://learn.microsoft.com/en-us/events/pdc-pdc08/es23
0:15:46 SMB2 Redirector
0:17:13 CopyFile API, 1MB I/O request
0:22:46 Application Guidelines


--

Regards,
Jones Syue | ???
QNAP Systems, Inc.

On 8/21/23 11:53, Jones Syue ??? via samba wrote:
>> OH - that's *really* interesting ! I wonder how it is
>> changing the SMB3+ redirector to do this ?
> 
> It looks like applications could do something and give a hint to SMB3+
> redirector, so far not quite sure how to make it,
> per process monitor (procmon) could show that write I/O size seems
> could be pass from the application layers,
well, sure, the application layer drives IO with the IO size it chooses. 
But when the application layer issues IOs larger then the maximum IO 
size allowed by SMB (8 MB iirc), the Windows SMB kernel client will 
segment the larger application segments into smaller ones.

-slow


---
Meet us at Storage Developer Conference (SDC)
On 18th to 21st September 2023 in Fremont, CA
More information at https://samba.plus/events

Meet us at the conference storage2day 2023!
26th & 27th September, in Frankfurt am Main
Event on Storage Networks & Data Management
Find more info at https://samba.plus/events

Ralph Boehme, Samba Team                      https://samba.org/
SerNet Samba Team Lead                     https://sernet.de/en/
SAMBA+ Samba packages                        https://samba.plus/
SAMBA+ Webinar                 https://samba.plus/samba-webinars


-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_signature
Type: application/pgp-signature
Size: 840 bytes
Desc: OpenPGP digital signature
URL:
<http://lists.samba.org/pipermail/samba/attachments/20230821/5620eae8/OpenPGP_signature.sig>

Hello everyone,

Sorry for the late reply. After doing some testing of the FastCopy 
utility with SMB3+ protocols using the 8MB I/O units you recommended 
Jones, we've managed to achieve ~400 MB/s reads and writes to local 
Windows clients and thus are already approaching the maximum I/O speed 
of the internal SSDs. This has already quadrupled our prior speeds so 
thank you ever so much for the recommendation!

We've also managed to get slightly faster speeds by selecting the
"Don't
wait for another FastCopy running to finish" option and we're currently
exploring how varying the FastCopy and smb.conf options on the client 
and server respectively are impacting performance as well.

Thank you ever so much for the references as well. I was unaware that 
Microsoft had a repository of previous conferences that were free 
accessible and contained talks regarding SMB on Windows, they look very 
informative!

Kindest regards,

Ivan

On 21/08/2023 10:53, Jones Syue ??? wrote:
> Hello Jeremy,
>
>> OH - that's *really* interesting ! I wonder how it is
>> changing the SMB3+ redirector to do this ?
> It looks like applications could do something and give a hint to SMB3+
> redirector, so far not quite sure how to make it,
> per process monitor (procmon) could show that write I/O size seems
> could be pass from the application layers,
> (https://learn.microsoft.com/en-us/sysinternals/downloads/procmon)
> this is a example for writing, using applications (Fastcopy.exe,
> explorer.exe, and robocopy.exe) to transfer a file named '1GB.img'
from
> local disk 'H:' to remote SMB mount disk 'Z',
> and monitor them with procmon:
>
> 1. FastCopy (FastCopy.exe), 8MB (need to revise FastCopy default setup)
> Process Name	Operation	Path	Detail
> System	TCP Send	10.19.7.63:64845 -> 10.19.250.98:445	Length: 8388724,
startime: 3459115, endtime: 3459149, seqnum: 0, connid: 0
> System	TCP Receive	10.19.7.63:64845 -> 10.19.250.98:445	Length: 84,
seqnum: 0, connid: 0
> FastCopy.exe	WriteFile	\\10.19.250.98\Public\1GB.img	Offset: 92,274,688,
Length: 8,388,608, Priority: Normal
>
> 2. File Explorer (explorer.exe), 1MB, drag-and-drop files.
> Process Name	Operation	Path	Detail
> explorer.exe	ReadFile	H:\1GB.img	Offset: 55,574,528, Length: 1,048,576,
Priority: Normal
> System	ReadFile	H:\1GB.img	Offset: 55,574,528, Length: 1,048,576, I/O
Flags: Non-cached, Paging I/O, Priority: Normal
> System	WriteFile	\\10.19.250.98\Public\1GB.img	Offset: 55,574,528, Length:
1,048,576, Priority: Normal
> System	TCP Send	10.19.7.63:64845 -> 10.19.250.98:445	Length: 1048692,
startime: 3545682, endtime: 3545689, seqnum: 0, connid: 0
> System	TCP Receive	10.19.7.63:64845 -> 10.19.250.98:445	Length: 84,
seqnum: 0, connid: 0
>
> 3. robocopy (robocopy.exe), 1MB, command 'robocopy.exe h: z: 1GB.img
/NP'
> Process Name	Operation	Path	Detail
> Robocopy.exe	ReadFile	H:\1GB.img	Offset: 520,093,696, Length: 1,048,576,
Priority: Normal
> System	WriteFile	\\10.19.250.98\Public\1GB.img	Offset: 520,093,696, Length:
1,048,576, Priority: Normal
> System	TCP Send	10.19.7.63:64845 -> 10.19.250.98:445	Length: 1048692,
startime: 3618938, endtime: 3618945, seqnum: 0, connid: 0
> System	TCP Receive	10.19.7.63:64845 -> 10.19.250.98:445	Length: 84,
seqnum: 0, connid: 0
>
> Perhaps i could do 'procmon' with other benchmark tool like
iometer,
> get back here later :)
>
> Found a video talking about SMB2 redirector and large I/O size,
> thouhd it is not up-to-date SMB3+ redirector still good to know about
> related design thoughts in the history:
> https://learn.microsoft.com/en-us/events/pdc-pdc08/es23
> 0:15:46 SMB2 Redirector
> 0:17:13 CopyFile API, 1MB I/O request
> 0:22:46 Application Guidelines
>
>
> --
>
> Regards,
> Jones Syue | ???
> QNAP Systems, Inc.
>
> --
> CAUTION: This email originated from outside of the LMB.
> Do not click links or open attachments unless you recognize the sender and
know the content is safe.
> .-jonessyue at qnap.com-.
>
-- 
Ivan Clayson
-----------------
Scientific Computing Officer
Room 2N249
Structural Studies
MRC Laboratory of Molecular Biology
Francis Crick Ave, Cambridge
CB2 0QH