SMB2FS OS3 port

[BSzili]

Thanks for testing! I think this puts the idea to bed that SMBv2 has high overhead

[alexh]

Thanks for this. I have been using the older Amiga 3.x SMBFS for many years.

[FlynnTheAvatar]

Thank you for porting smb2fs handler to AmigaOS 3.x.

I tried it yesterday on my A500 with ACA500+, X-Surf 500 and an ACA1221lc with Amiga OS 3.2.1. Worked flawlessly for the basic stuff (copying stuff to and from my NAS).

[BSzili]

That's great to hear. Currently I'm preparing a new release for filesysbox.library that includes an installer and both 68k and 020 versions. It won't likely change much until I manage to implement ACTION_DIE properly.

[patrik]

Originally Posted by BSzili

Thanks for testing! I think this puts the idea to bed that SMBv2 has high overhead

Ran the same test with smbfs 1.74 from aminet to see how they differ:
A3000-X-Surf-100-smbfs.txt

The biggest difference seems to be the fixed cost for each operation which appears to be much lower in smbfs.

smb2-handler:

Testing directory manipulation speed.
File Create:           13 files/sec  |  CPU Available: 0%
File Open:             14 files/sec  |  CPU Available: 0%
Directory Scan:       271 files/sec  |  CPU Available: 1%
File Delete:           14 files/sec  |  CPU Available: 0%

Seek/Read:             31 seeks/sec  |  CPU Available: 0%

Testing with a 512 byte, MEMF_FAST, LONG-aligned buffer.
Create file:        32750 bytes/sec  |  CPU Available: 0%
Write to file:      33664 bytes/sec  |  CPU Available: 0%
Read from file:     29568 bytes/sec  |  CPU Available: 0%

smbfs:

Testing directory manipulation speed.
File Create:           51 files/sec  |  CPU Available: 1%
File Open:            151 files/sec  |  CPU Available: 0%
Directory Scan:       293 files/sec  |  CPU Available: 20%
File Delete:           90 files/sec  |  CPU Available: 0%

Seek/Read:            105 seeks/sec  |  CPU Available: 0%

Testing with a 512 byte, MEMF_FAST, LONG-aligned buffer.
Create file:       105280 bytes/sec  |  CPU Available: 2%
Write to file:     109312 bytes/sec  |  CPU Available: 0%
Read from file:    113280 bytes/sec  |  CPU Available: 0%

Also, this would also be indicated by smbfs reaching maximum throughput at smaller reads.

smb2-handler:

Testing with a 32768 byte, MEMF_FAST, LONG-aligned buffer.
Create file:       288651 bytes/sec  |  CPU Available: 0%
Write to file:     332546 bytes/sec  |  CPU Available: 0%
Read from file:    499712 bytes/sec  |  CPU Available: 0%

smbfs:

Testing with a 32768 byte, MEMF_FAST, LONG-aligned buffer.
Create file:       105966 bytes/sec  |  CPU Available: 42%
Write to file:     111890 bytes/sec  |  CPU Available: 41%
Read from file:    626688 bytes/sec  |  CPU Available: 0%

One thing to note here too is that smbfs gets a "CPU Available: " above zero for some tests, for example create and write 32kB, while it should really get 0% and much higher result - the cpu time is wasted waiting on something in those cases.

If compared to smb2-handler, it always has 0% "CPU Available", which makes complete sense in this scenario, while smbfs is up to some schenanigans causing some delays somewhere where the CPU can idle.

What I mean with that is that on this machine, the CPU is the bottleneck, so it should never be waiting in a test like this as:
- No DMA involved
- The network card at 100Mbit is way faster than what the CPU can feed data through the TCP/IP stack (~670kByte/s down, ~400kByte/s up)
- The server should reasonably not cause any waiting

This is usually the case on the Amiga, unless you have a 060 and a slow network interface so that becomes the bottleneck. Also important to consider is that the more CPU the application uses for something, the less CPU will be available for the TCP/IP-stack and sana2-driver to handle the data, so in general any CPU saved in the application results in a higher effective transfer speed.

So a lot of babbling. What I want to say is that the maximum throughput achieved at large reads/writes can likely not be improved much in smb2-handler, but the speed of small operations like open file and small buffer reads can probably be improved substantially if the fixed CPU cost for each operation can be lowered.

[steenl]

@BSzili Great work! Works like a charm on my A4K with PCI ethernet and is now my preferred way to connect to my NAS.

Thank you!

[BSzili]

Thanks everyone, props go to salass00 for creating the handler! Let me know if you find any issues.

Originally Posted by patrik

So a lot of babbling. What I want to say is that the maximum throughput achieved at large reads/writes can likely not be improved much in smb2-handler, but the speed of small operations like open file and small buffer reads can probably be improved substantially if the fixed CPU cost for each operation can be lowered.

It could probably be improved by getting rid of some dynamic memory allocation. I'll test this with one of the functions to see if it makes any difference.

[DisasterIncarna]

great work, i've been using my asustor nas as a middleman for all my smb shares, it connects to any linux/windows shares as well as providing its own as that nas lets me use/support SMB1, wit this it looks like i wont need to enable SMB1 support as the amiga side was the only reason i had to do this.

[mfilos]

@BSzili
Great work! It works amazingly on my Vampire V4SA. Awesome stuff \o/

[patrik]

Originally Posted by BSzili

It could probably be improved by getting rid of some dynamic memory allocation. I'll test this with one of the functions to see if it makes any difference.

Exciting

[BSzili]

False alarm. I overestimated how much impact dynamic memory allocation has in proportion to the rest.

[amigafreak68k]

@BSzili

is it possible, to implement a option for a volume name?

Actually I get "fritz.box-fritz.nas" as volume name, but cooler would be just "Fritzbox"...

[mfilos]

Originally Posted by amigafreak68k

@BSzili

is it possible, to implement a option for a volume name?

Actually I get "fritz.box-fritz.nas" as volume name, but cooler would be just "Fritzbox"...

That would be nice indeed
I shared a folder on my Win11 PC named shared and I get a:
<ip address>-shared drawer in my Workbench (I put a Disk.info as well).

[FlynnTheAvatar]

Originally Posted by mfilos

That would be nice indeed
I shared a folder on my Win11 PC named shared and I get a:
<ip address>-shared drawer in my Workbench (I put a Disk.info as well).

Yes, there is a "VOLUME" option. Just add it with the desired volume name after the URL.

[yelworC]

I tried this today and it simply works

Thank you for porting this to our good old classic OS3

[amigafreak68k]

Another thing:

If I would copy files with special characters (like locales for french and spain), then I get AmigaDOS error 210. With SMBFS I remember I could cooy these files. Are there a option to get this fixed?

[DJBase]

Installed on my A1200 and used it for a few days now. Wrote a few scripts for mount&dismount and added to the DOpus Magellan menu.

Having troubles with special characters in password and filenames it seems. So looking forward for a new version for testing.

[amigafreak68k]

As I said

[BSzili]

Originally Posted by amigafreak68k

Another thing:

If I would copy files with special characters (like locales for french and spain), then I get AmigaDOS error 210. With SMBFS I remember I could cooy these files. Are there a option to get this fixed?

Originally Posted by DJBase

Installed on my A1200 and used it for a few days now. Wrote a few scripts for mount&dismount and added to the DOpus Magellan menu.

Having troubles with special characters in password and filenames it seems. So looking forward for a new version for testing.

I haven't tested any files or passwords with international characters yet. If possible, could you give me any examples of file names you had problems with?

[tygre]

Hi all!

Originally Posted by amigafreak68k

Another thing:

If I would copy files with special characters (like locales for french and spain), then I get AmigaDOS error 210. With SMBFS I remember I could cooy these files. Are there a option to get this fixed?

I had this question too: SMBFS handled ASCII 256 and I had contributed some code to handle UTF-8 characters (very crude, but enough to let me play MP3 with UTF-8 names via my network).

Can SMBFS2 handle such files?

Cheers!