Reply
Thread Tools
Posts: 2,152 | Thanked: 1,490 times | Joined on Jan 2006 @ Czech Republic
#21
Originally Posted by luca View Post
My opinion is that it's either too badly written that they're ashamed of showing it, or the device is so crappy that they're ashamed of showing it. However usually it's both
Usually it is due to licencing terms. But still it may be crappy :-)

Some parts are closed, the glue between kernel and proprietary parts is open (mainly due to GPL requirements), see some details here http://garage.maemo.org/pipermail/cx...er/thread.html

As for cx3110x (the glue) - its quality is not exactly stellar, 770 version being much worse with CPU. For N800 version CPU consumption it the open part should be fine/better but it is hard to say without benchmarks. The wi-fi chip is connected over SPI bus, OMAP2 has SPI interface which handles SPI communication without needing much CPU, 770 uses McBSP chip (serial port) for SPI and busy-loops waiting for each result from the chip which eats a lot of CPU.
__________________
Newbies click here before posting. Thanks.

If you really need to PM me with troubleshooting question please consider posting it to the forum instead. It is OK to PM me a link to such post then. Thank you.
 
Posts: 25 | Thanked: 24 times | Joined on Mar 2006
#22
This PSM tip has resulted in much improved browsing performance for me, thanks!
 
brontide's Avatar
Posts: 868 | Thanked: 474 times | Joined on Oct 2007 @ Capital District, NY, USA
#23
Has there been any movement on the WiFi performance? Developer list said they have gotten the 770 up to 1.2mB/s download which would be 2 or 3 times the download performance I'm seeing on my 810 ( Usually 500kB/s over 802.11g ).

I've tried tuning the kernel, turning energy savings on and off, and many other tricks without any change. There is some other ceiling that I'm hitting and I have no idea where it is.

I think this is the problem has been discovered, but no word on if the n8x0 has actually gotten the patches to prevent network busylooping

https://garage.maemo.org/pipermail/c...ry/000012.html

A while ago I looked for various kernel docs to see what's happening in the
wlan driver and what can be done to reduce cpu load. My impression was that
tasklet can be only preempted by hardware interrupts, so it is impossible to
sleep in it and give cpu resources to userland applications. If that is true,
no matter if n800 driver looks nicer, it must end up busylooping too.

Though on Nokia 770 cpu usage is attributed to the application doing (for
example wget) and on N800 it is attributed to 'OMAP McSPI/0' process.
So no matter how much I try in userland I will always be hamstring by bad programming in the kernel.

Last edited by brontide; 2008-03-13 at 14:57.
 
Posts: 55 | Thanked: 6 times | Joined on Feb 2008
#24
no, Linux kernel's fine, it's the nokia driver and how they've implemented their wifi device power saving mode that's causing the poor performance. Setting IT's PSM to intermediate or none really improved web browsing, combined with brontide's optimization tips.
__________________
**************************************

My N800: OS2008, Fanoush's initfs, SanDisk 4Gig x 2 SDHC, BoxWave FlexSkin and Sync Charger, Homebrew External Power Pack (4xAA NiMH Rechargeable 2400 mA)

Greetings from Sunny San Diego, California!
 
brontide's Avatar
Posts: 868 | Thanked: 474 times | Joined on Oct 2007 @ Capital District, NY, USA
#25
Yea, my tips would be a whole lot faster if we can break this speed wall on the wifi. Maybe even allow a few more connections without swamping the cpu.

I'm not expecting to peak at over 200MB/s like some of my work systems, but being able to clear 1MB/s without consuming 100% of the CPU would be nice. The wifi chipset is another whole arm chip! you think they could do something with that extra power.
 
Posts: 503 | Thanked: 267 times | Joined on Jul 2006 @ Helsinki
#26
Originally Posted by brontide View Post
Has there been any movement on the WiFi performance? Developer list said they have gotten the 770 up to 1.2mB/s download which would be 2 or 3 times the download performance I'm seeing on my 810 ( Usually 500kB/s over 802.11g ).
Performance depends on the way you are downloading data to the device. That ~1.2MB/s was measured as pure download speed, even without storing the received data on memory card. Saving data to memory card (that's natural for normal use) will drop the performance to ~1MB/s, using scp will decrease it even more because of extra load on cpu introduced by encryption. Poorly written download client may screw up the performance too. What have you used in your tests?

I've tried tuning the kernel, turning energy savings on and off, and many other tricks without any change. There is some other ceiling that I'm hitting and I have no idea where it is.

I think this is the problem has been discovered, but no word on if the n8x0 has actually gotten the patches to prevent network busylooping

https://garage.maemo.org/pipermail/c...ry/000012.html
Have you checked followup messages in the cx3110x-devel mailing list? It was later explained that N800/N810 wlan driver does not do busylooping anymore and can sleep while DMA transfer is in process, giving cpu to other tasks. That does not improve peak performance, but makes it more friendly to other applications. So the current N800/N810 driver already is an improvement over the default driver from 770.

If you are interested in getting answers to your wlan driver related questions, it is better to subscribe to cx3110x-devel mailing list and post to it. Kalle Valo prefers to keep the discussion there (he mentioned that several times) and he can easily miss this thread on ITT.

So no matter how much I try in userland I will always be hamstring by bad programming in the kernel.
Cheer up The wlan driver does get improvements over time. If you can't wait, you or anybody else can also hack the driver yourself.

There is still a lot of potential in improving performance of both Nokia 770 and N800/N810 driver.
 

The Following User Says Thank You to Serge For This Useful Post:
brontide's Avatar
Posts: 868 | Thanked: 474 times | Joined on Oct 2007 @ Capital District, NY, USA
#27
Originally Posted by Serge View Post
Performance depends on the way you are downloading data to the device. That ~1.2MB/s was measured as pure download speed, even without storing the received data on memory card. Saving data to memory card (that's natural for normal use) will drop the performance to ~1MB/s, using scp will decrease it even more because of extra load on cpu introduced by encryption. Poorly written download client may screw up the performance too. What have you used in your tests?
time wget -q -O /dev/null http://a.b.com/10mbrandomdata

With my best tuning it runs into a hard wall ( where nothing you throw at it makes a difference ) at around 5000kbps / 625kBps. Downloading to memory is actually the way the browser works, so it's not an unfair test.
 
Reply


 
Forum Jump


All times are GMT. The time now is 23:15.