Prev: Routing between 2 NICs
Next: Jumbo frames and compatibility with NAT
From: Fguy on 9 Feb 2010 15:05
Greetings. I am have Debian Lenny on a wireless home network that also
has a couple of Windows XP machines. They connect via a linksys
router.

I am finding file transfers between windows and linux samba share,
both directions, very slow. This slowness also applies to doing
Windows to linux file transfers using sftp. It takes about 4 times
longer to transfer a 10 MB file to Linux than it does another Windows
XP box.

Anyways, I could use a little guidance to give some focus to my
efforts. I don't mind doing the legwork, but I'm not sure which
direction. If it was just samba, then there is a lot of info on the
net, but things aren't so clear when it applies to sftp as well.

Thanks in advance.

p.s. my Linux box is an older machine, a PIII 733 with 512 MB of RAM.
All unneeded services that I have installed are disabled, just ssh and
samba is running.

From: Rick Jones on 9 Feb 2010 15:31
You should first verify that basic network performance between your
Linux and Windows systems is good. Naturally I'd suggest netperf :)
there are though other tools.

If the shares are serviced via request/response, don't rely simply on
a netperf TCP_STREAM test - some variation on a netperf TCP_RR test
would be in order.

happy benchmarking,

rick jones
http://www.netperf.org/
--
the road to hell is paved with business decisions...
these opinions are mine, all mine; HP might not want them anyway... :)
feel free to post, OR email to rick.jones2 in hp.com but NOT BOTH...
From: Fguy on 9 Feb 2010 15:34
On Feb 9, 3:31 pm, Rick Jones <rick.jon...(a)hp.com> wrote:
> You should first verify that basic network performance between your
> Linux and Windows systems is good.  Naturally I'd suggest netperf :)
> there are though other tools.
>
> If the shares are serviced via request/response, don't rely simply on
> a netperf TCP_STREAM test - some variation on a netperf TCP_RR test
> would be in order.
>
> happy benchmarking,
>
> rick joneshttp://www.netperf.org/
> --
> the road to hell is paved with business decisions...
> these opinions are mine, all mine; HP might not want them anyway... :)
> feel free to post, OR email to rick.jones2 in hp.com  but NOT BOTH...

OK thanks I can work with that, though it might take me a while to
learn some networking basics,. I had assumed that since the problem
existed for two different applications, samba and sftp, that that more
or less implied that basic network performance was in fact bad, in
which case a netperf would just confirm what I already knew. Perhaps I
cant make these kind of assumptions?
From: Fguy on 10 Feb 2010 08:45
OK Rick, much appreciated, It's going to take me a little while to
work through this stuff and put it to the test. It'll be a good
learning experience. I'll let you know how it goes, but it might be a
week or two.
From: Fguy on 11 Mar 2010 09:53
ok Rick, or anyone else who is following the action, I finally got
around to trying a few more things. Performance testing is not my cup
of tea though, I just don't have the disciplined mind set to do it
properly. I ended up blowing away and resinstalling Lenny 5.0.4 on a
PIII 733 with 512 MB RAM. It's just a spare computer for
experimentation. I am focusing only on FTP and HTTP for now.

tests are done using an 11.9 MB zip file. None of the tests showed
significant CPU usage on either XP or Linux.

download from my ISP to Windows XP via ftp - 23 sec.
download from Linux to XP via ftp - 30 sec

download from ISP to XP via HTTP - 10 sec
download from Linux to XP via HTTP 20 sec

Using the native SystemMonitor applet in Linux, the Network History
section, which seems to measure throughput, showed quite a bit of
oscillation as the file was being transferred by HTTP.

download from my ISP to Linux using http was fast, no problem.

Hopefully this info might provide a little direction. My tests could
be improved upon I'm sure, but that's the best I can do for now. At
least can we agree that there is a problem?

regards.
 | 
Pages: 1
Prev: Routing between 2 NICs
Next: Jumbo frames and compatibility with NAT