On Fri, 2002-08-16 at 13:27, Kevin Geiss wrote:
> well, first of all, NFS will preserve user and group ownership of the
> files, as well as ALL the file attributes. with samba, you only get the
> owner and the time stamp, I don't think you get any group attributes.
Indeed. This isn't always a good thing, though. I had a public directory
that was to be accessed by several Linux boxes and a few windows boxes, all
with various users. NFS kept insisting on preserving the user on the
files, though, so at any given time, most of the users couldn't access some
of them. With SMB, you can guarantee that everybody that can access the
share can work with the files on it.
> For performance, all I can offer is a testimonial. I was about to
> reinstall windows 98, back in the bad old days when I still had a
> windows machine at all. I had set up samba to do drive share. I ended
> up with a large .zip file I wanted to save before reinstalling, so I
> started copying it to one of the drive shares that were on my linux
> machine. The file was big, I think a couple hundred megabytes. I
> started doing other stuff, then after a while started wondering why it
> was still copying. I watched the rate the data was getting to my linux
> machine, then calculated it would take 10 hours!! so then I installed
> an ftp client and ftp'd the file in 10 minutes (i have a 100mb
> ethernet). granted, I'm no samba expert, so perhaps the samba setup on
> my linux box wasn't optimal.
I'd go with the non-optimal setup scenerio :-) I know what you mean,
though. I have one Linux box with samba that always freezes when you try
to copy over large files from a Windows box. On the rest of my systems, it
works fine. Same version of samba, same config. Very weird.
Anyway, I just did some seat-of-my-pants tests with a roughly 100M file
copied from a Linux box to a "virtual" user-mode server running both NFS
and samba on the same directory. I did several copies back and forth using
both protocols. I got roughly 1.6MBs (bytes, not bits) for NFS and 2.1MBs
for samba. Keep in mind that the figures are a bit slow since I/O on the
file-based user-mode is inherently slow. In any event, far from being a
lot slower than NFS, it seemed to be noticably faster!
This is why I am still in a search for "real" benchmarks. I refuse to think
that nobody has done them yet I can't find 'em anywhere.
--
Kurt Granroth - "KDE -- Conquer Your Desktop"
KDE Developer/Evangelist |
granroth@kde.org
http://www.granroth.org |
kurt@granroth.org
(text/plain)