S0urcerr0r | # 1 | 2010-05-28, 02:18 | Report

Does anyone know of any fremantle (or debian) based test suites (ARM compatible) that can benchmark all or some of the following hardware components of our N900's:

* 256MB Mobile DDR: Speed and/or Reliability tests

* BUS: How much data bandwidth there is available to use between the different hardware components at the same time (SGX535, WLAN, Mass storage devices, Sound) - maybe this isnt strictly related to the RAM speed?

* NAND & eMMC: Regular SSD speed tests - sequential & single sector reads / writes

* Micro SD: Same speed tests as for NAND... it may be interesting as well to compare BUS-hogging between NAND, eMMC and Micro SD.



The reason that i find these tests interresting is to find out why some tasks on the N900 causes heavy overload.

Among these tasks i think of the bittorrent clients QtRapids and Transmission that really hogs the N900... they max out at 150-200 KB/s and slows down the system plentyful. this happens both with 3G and WLAN... if the reason is that the eMMC gets IO overloads or just cant handle that speed with small random sector writes it would be interresting to know if the Micro SD slot could be a candidate for handling such tasks, or if it atleast would prevent slowdowns for the system... of course it may also be that the torrent clients arent code optimized...
another interresting aspect would be if it would be possible to run SSD cache software that puts some of the heavy flash load in RAM and make more efficient use of the NAND/eMMC. Similar software solutions have been made for netbooks with weak SSD's but of course disk cache software isnt 100% data reliable in case of crashes or power outage.

these tests may also be useful to see if there is any relation between the SGX535 V-Sync problems and the BUS... maybe the bus isnt wide enough, or just cant handle enough data throughput. The Sony Vaio P netbooks also use the SGX535 and have a high resolution of 1600x768 (in 24 or 32 bit color?), so there should be enough horsepower to handle the N900's 800x480 without frameskips and V-Sync problems, if there is a proper framebuffer in ram. Iphone 3GS (SGX) dont suffer from these V-Sync problems but on the other hand it only have 480x320 resolution ...maybe this issue only is a matter of drivers.

S0urcerr0r | # 2 | 2010-05-28, 12:27 | Report

i did one test today for USB speed comparisons. not a very proper and reliable one though:

i reflashed PR1.2 to see what USB speeds can be achieved while using flasher3.5... i booted the n900 in flasher mode (holding u+connecting the usb cable) ...excellent mode for testing i think - since its basically the cleanest environment to test max hardware speeds.

i averaged at the speed 13602 KB/s while continously writing the 170mb flash file... the smaller flashes bursted at higher rates though .all according to flasher3.5's verbose text output.

in comparison to that i never reach speeds above 6000KB/s when connecting the N900 in mass storage mode (while maemo is running in the background). it should be caused by one of these reasons:
1. maemo os hogging the system so much that the overall system speed suffers
2. while running maemo, BUS bandwidth is overloaded with IO tasks to other hardware components.
3. less optimized drivers in maemo.
4. unknown reason?

ps. actually im not sure if it was the N900's, or the laptops usb speed that got maxed out during the test. i should have tested with a max spec USB2 port on another computer.

---

Last edited by S0urcerr0r; 2010-05-28 at 12:30.

Tony Grunt | # 3 | 2010-05-28, 12:44 | Report

Did a 1Gig file copy and it averaged 17500-17600 KB/sec.
My 16gig Class 6 microsd did 7000-7100 KB/sec for the same file.

---

Last edited by Tony Grunt; 2010-05-28 at 12:47.

zimon | # 4 | 2010-05-28, 14:57 | Report

If hdparm would be ported to N900, then could be easily tested with -t and -T options.

With dd, basic read/write tests can be already made.

These kind of tests could be part of the HealthCheck program tools.

mackan1 | # 5 | 2010-05-28, 15:03 | Report

What would happen if you can change 256MB Mobile DDR to 512MB?

would be fun to see if there would be any difference. that's if it has a memory chip

noobmonkey | # 6 | 2010-05-28, 15:48 | Report

yup! its on my very very long todo's!! happy to accept help where needed!

S0urcerr0r | # 7 | 2010-05-29, 01:28 | Report

awesome! thanks for all the answers!
i look forward to the additional features of healthcheck

it would also be nice if someone made a clean environment image that can be flashed to another partition. stripped of everything hogging system performance (GUI and background tasks). only for testing purposes like ram checks/benchmark and usb, NAND and eMMC benchmarks. it seems that maemo is quite a hogger even though its lightweight, but i guess its because its a full blown Linux OS.
i remember the days when people used DOS+Windows 3.1... on a "33MHZ+4MB ram" computer, Win3.1 started in a half second. not a very pretty GUI though but 100% snappy because there was almost no background tasks ...i guess GUI's for linux were even faster than that, back in those days.

unfortunatly im not experienced enough to create a benchmark image for the N900... ill try learning it though when i get access to a desktop. have been using the N900 as a main computer for months.

EDIT: ill edit the first post if any benchmark apps comes around (or health check gets add features) and can do tests like these.



nice speeds! can u please benchmark the torrent clients QtRapids and Transmission.

Compare OS responsiveness while downloading in highest possible speed - one test for the eMMC, and another test for the MicroSD?
if we're lucky the system stays responsive while downloading to the microsd. i have no microsd yet so i cant do test right now.

---

Last edited by S0urcerr0r; 2010-05-29 at 01:43.

clovis86 | # 8 | 2010-09-13, 10:22 | Report

interested for a class 10 bench .
bought a 32 Gb a month ago ( class 2 )
and sold it a week ago coz was too slow.
I know that it's linked to the class but
is a class 10 for example will be used at 100% on the N900 ?
regarding theorical read / write speed

Would change a lot for nitdroid on it

shadowjk | # 9 | 2010-09-13, 15:55 | Report

The class speeds are only achievable with sequential straming writes. A OS disk isn't used that way.

Still if it's class 10 because it's actually 5 times faster than class 2, and not just 5 class 2 flash chips in a multichannel configuration, it shold be faster for os use too.

clovis86 | # 10 | 2010-09-13, 15:57 | Report

ok si class 10 will definitely be faster for storage under maemo and / or os hosting too ^^
thanks