[gentoo-user] Re: USB mass storage device slow in Gentoo, fast in Windows...? [SOLVED]

[Paul Hartman]

On Sat, Dec 26, 2009 at 4:54 PM, Paul Hartman
<paul.hart...@gmail.com> wrote:
> Hi,
>
> I got a Nokia N900 linux internet tablet/phone a few days ago, and
> when I connect it in USB Mass Storage mode to a Windows Vista computer
> I can write at 17MB/sec, but when I connect it to my Gentoo box my
> writes are really slow, between 500-900kb/sec depending on if I mount
> in "sync" mode or not. As far as I know it should be just a totally
> standard/generic mass storage device. (there were no drivers or
> software install needed in windows, it just worked)
>
> Other USB devices plugged into the same port go full speed, and AFAIK
> everything appears as if it should be high speed USB 2.0. Has anyone
> seen something like this before? I'm not sure what the deal is. It
> takes 20 minutes to copy 1 gigabyte from Linux and takes just under 1
> minute to do the same in Windows.
>
> I'm not sure about debugging USB or what the options are. Everything
> I've used previously has worked without any hassle.

Solved. The problem was CFQ I/O scheduler. It was several times slower
than the others, for whatever reason.

Here is the scoreboard:

single-file: 1m25s

multi-file (same total size):
cfq: 6m51.439s
noop: 3m0.733s
anticipatory: 1m44.348s
deadline: 1m36.804s

So, the winner is deadline. CFQ doesn't make it to the podium. :)

Thanks,
Paul

[Mick]

Hmmm ... reading at the help files I thought that CFQ was the default/best
option for a desktop. Is there such a thing as a best fit here?
--
Regards,
Mick

[Alan McKinnon]

On Tuesday 05 January 2010 10:15:00 Stroller wrote:
> On 5 Jan 2010, at 06:21, Mick wrote:
> >> ...

> >> Solved. The problem was CFQ I/O scheduler. It was several times slower
> >> than the others, for whatever reason.

> >> ...

> >
> > Hmmm ... reading at the help files I thought that CFQ was the
> > default/best option for a desktop. Is there such a thing as a best fit
> > here?
>

> I had this notion that it was the worst option for everything, unless your
> computer has 96+ CPUs.

I had this notion that the author of the help files was talking through a hole
in his butt and just expressing his own views.

--
alan dot mckinnon at gmail dot com

[Stroller]

On 5 Jan 2010, at 06:21, Mick wrote:
>> ...

>> Solved. The problem was CFQ I/O scheduler. It was several times slower
>> than the others, for whatever reason.

>> ...

>
> Hmmm ... reading at the help files I thought that CFQ was the default/best
> option for a desktop. Is there such a thing as a best fit here?

I had this notion that it was the worst option for everything, unless your computer has 96+ CPUs.

Stroller.

[Mick]

2010/1/5 Alan McKinnon <alan.m...@gmail.com>:

Ha, ha!

What does experience show to be a best option for a desktop that has:

a) Single CPU?
b) Dual core?
c) Quad core?

--
Regards,
Mick

[Szénási István]

And what about the BFS scheduler? I know, that it isn't in the
mainline kernel, bit I've heard a lot of good about that.
If you send me the size and the number of the test files, I'll make an
other benchmark with the CFQ, the Deadline and the BFS scheduler on a
Dual Core machine. :-)

--
Szénási István

[Stroller]

On 5 Jan 2010, at 11:39, Mick wrote:
> What does experience show to be a best option for a desktop that has:
>
> a) Single CPU?
> b) Dual core?
> c) Quad core?

On 5 Jan 2010, at 12:38, Szénási István wrote:
> And what about the BFS scheduler? I know, that it isn't in the

> mainline kernel, bit I've heard a lot of good about that. ...

I didn't think we're supposed to talk about that. :-/

Stroller.

[James Ausmus]

2010/1/5 Szénási István <sze...@gmail.com>

BFS is a CPU scheduler, CFQ, Deadline, No-Op, and Anticipatory are all I/O schedulers...

-James
 

--
Szénási István

[Szénási István]

> BFS is a CPU scheduler, CFQ, Deadline, No-Op, and Anticipatory are all I/O
> schedulers...

Sorry, I've realized that when I googled for benchmarks and it was
compiled against the CFS scheduler. :-)

--
Szénási István