[Arm-netbook] 9nov11: progress on allwinner A10

[Gordan Bobic]

On 11/13/2011 08:07 AM, Michelle Konzack wrote:
> Hello Gordan Bobic,
>
> Am 2011-11-12 21:52:42, hacktest Du folgendes herunter:
>>   If it is 3x faster than SD cards that still makes it about 300x slower
>>   than a proper SSD.
>
> Even SSD is only cooking with NAND Flash.
>
> However, my NAND Flash (Micron) makes arround 120 MByte/sec
>
>>> 1)  NAND is good for the Operating system to boot up or load programs
>>> 2)  SDxx cards are good for storage and always changing data
>>
>>   For an embedded system where the OS doesn't change often, sure. But if
>>   you are running Fedora and you do yum update you might as well leave it
>>   overnight. Same goes for Ubuntu and apt. I am still unconvinced that the
>>   W and $ budget wouldn't be better spent elsewhere...
>
> Ehm, my two prototype PanePC Workstations are a Marvel Discovery MV78200
> with 1200 MHz and Marvel Armada 300 with 2000 MHz.
>
> Both computers are booting and starting programs much faster then my AMD
> Athlon XP 2800+ with 3 GByte of memory and 2"5 7200 RpM SATA drive.
>
> Debian Squeeze needs around 8 Seconds to Boot-Up into WDM.
>
> The third try will be a Marvel Armada 1000.
>
> Operating System is Debian GNU/Linux Squeeze/ARMEL.  Nothing embedded.

I'm not talking about _LOADING_ speeds. That's easy. That's just read 
IOPS. My SheevaPlug boots faster off the SD card than my 3.2 GHz quad 
Core2 from a 7200 rpm disk. The problem is that usage != read-only. It 
is the write IOPS that are important, and anything short of a good SSD 
(nowdays only commonly available in SATA form) performs attrociously. 
Things like package installing/updating is particularly bad, as the 
package database gets written during the process in a transactionally 
safe and thus fsync()-heavy way.

You can have great boot speeds and still have an overall user-experience 
that sucks due to slow randon-write IOPS.

Gordan