Samba IRC

<mkl_> hi...
does any one know that is there any implementation trying to make writing file to samba completely in kernel?
like reading files from samba, with sendfile and mmap.....
My cpu and memory copy is so slow that became the bottleneck, and I would like the writing to samba not going to user space if possible....

<Berge> mkl_: No, there is no Samba implementation in kernel space.
mkl_: If context switching is a major concern on your platform, I'm not quite sure if you want to run Samba on it.

<mkl_> is there any substitution? I would need samba to share files....
but writing to samba is to slow...

<Berge> If your Samba installation performs poorly, there might be other causes than context switching.
CIFS is not a very efficient protocol, especially not with Windows clients.
(In my experience, at least.)

<Berge> What architecture is this?

<mkl_> my target is to improve the write performance...
<Berge> Have you looked at http://www.samba.org/samba/docs/man/Samba-HOWTO-Collection/speed.html ?

<mkl_> faraday FA526

--> hi365 (n=hi365@bzq-219-141-66.static.bezeqint.net) has joined #samba
tarbo (n=me@unaffiliated/tarbo) has joined #samba
<hi365> is there some way to reread the configs wothout stoping smbd?

<mkl_> yes, I've read them....

<-- tarbo has quit (Client Quit)
--> tarbo (n=me@unaffiliated/tarbo) has joined #samba

<mkl_> the read performance is improved, but I found no way to improve the write performance.........

<mkl_> btw, I can't subscribe any mailing list from https://lists.samba.org/mailman/, neither can I send mail to request@lists.samba.org......
what's up to the mailing list?

<Berge> mkl_: You'd have to ask the mailing list admins for that, I'd guess.

<mkl> Berge: why the writing to samba have to go through user-space? could it be implemented in kernel space, like sendfile?

---