So, when the N900 is connected, then turned on, it detects the USB connection to be
and when instead the N900 is turned on, then connected, it detects the USB connection to be
In the first case it goes into OHCI (USB 1.1) mode and full-speed, i.e. 12Mb/s, and the second case it goes into EHCI (USB 2.0) mode and high-speed (480Mb/s). No wonder there's a speed difference!

Now the question is why it does this. But often USB behaves a bit better when all devices are switched on before connecting anything, in my experience.

EDIT: Those dmesg outputs are from the laptop, not the N900? Not that it matters much though.

The dmesg output is from the laptop.
(I never thought of running dmesg on n900!!!)

Windows xp sp3.
I have connected device extensively now about 30 times (did some trial&error video encoding, had to upload video to device to see if it works with current encoder settings) and it looks like it's sometimes recognized as usb 1.0. Windows gives warning and write speeds are low. Totally random :| I do not know how to reproduce this. Additionally sometimes n900 doesn' t recognize connection on first time. Just good old plug off plug in works most of time. Doesn't feel very reliable.

How you can make bug raport when you cant reproduce bug?

Guys, I think I have solved the problem (for me anyway)! It seems to depend on the I/O scheduler you've selected. I was using CFQ. I just did some tests with the various schedulers and using the deadline scheduler gave me nearly the same speed for multi-file copy as it did for single-file copy -- around 17M/s.

You can change the scheduler like:

echo deadline > /sys/block/sdg/queue/scheduler

(change sdg with your device name)

For the record, all were the same speed in single-file copies, but for multi-file these were the times in my test:

cfq: 6m51.439s
noop: 3m0.733s
anticipatory: 1m44.348s
deadline: 1m36.804s

I tend to always only build a single scheduler... deadline

CFQ is trying to be fair and give each process a slice of time acessing the disk, but flash storage hates random writes and loves sequential writes.

My guess is that deadline puts more emphasis on writiing out those blocks that are close to eachother, regardless of which process wrote them.

On a possibly related note, I just copied 3GB of data to the sd card (mmcb1k1p1 /media/mmc1) over usb in a couple of minutes, then disconnected the usb and moved them from the memory stick to MyDocs/.documents and its been going for two hours. It's definitely doing it as its slowing down the rest of the system and the files are slowly appearing in the target folder.

I cancelled the mv and its only copied 1.4GB of it. The IO is still churning though the CPU time in top. Maybe a courtesy restart is in order

I've also seen really slow speeds at internally N900-to-N900 operations. For example I tarred up about 1gig of small files, transferred to N900 (MyDocs), then untarred it from xterm on the phone. It took tens of minutes and the UI of phone was completely unusable while this was happening. (even pressing the unlock switch would take several minutes to respond). I wonder if there's ionice for N900 somewhere?

Hello There i think this is an old thread any way i have a problem with h-e-n after using it with usb flash
the transfer speed between n900 and the pc goes down from 15mb/s to 4mb/s
but if i rebooted my n900 the speed goes between n900 and pc goes again to 15mb/s
and if i used h-e-n the speed goes back to 4 mb/s any solutions

It seems that the same thing also happens with me, but reboot doesn't help.
I've got a fast class10 UHS-1 microSD 16GB, by Samsung. Originally, speed beetween PC and MicroSD in N900 through USB was 10MB/s. Currently it's only 3MB/s, and copying on the phone is also max 4,5MB/s. And the graphical interface lags like hell.

Nokia N900 with CSSU-testing and CSSU-thumb installed, except the last update, I'm afraid that hildon update brick my device.

What can I do guys? Please help.