Hi,
I know this is a bit off-topic, but I've recently bought a 256GB SD card with the following properties:
name: SD256 manfid: 0x000027 oemid: 0x5048 date: 05/2015 preferred_erase_size: 67108864
The device seems to support random access for up to 5 allocation units, judging from the erease-size I guess this is a TLC device. All in all the controller performs rather miserably when used as root-filesystem - I am actually a bit surprised such a weak controller is paired with flash amounting to > 40-50$.
As I have several cards with manfid 0x27 and phison controllers, I started to wonder which company is actually behind this 0x27 id?
Thank you in advance, Clemens
On Tue, Apr 11, 2017 at 2:48 PM, Clemens Eisserer linuxhippy@gmail.com wrote:
Hi,
I know this is a bit off-topic, but I've recently bought a 256GB SD card with the following properties:
name: SD256 manfid: 0x000027 oemid: 0x5048 date: 05/2015 preferred_erase_size: 67108864
The device seems to support random access for up to 5 allocation units, judging from the erease-size I guess this is a TLC device. All in all the controller performs rather miserably when used as root-filesystem - I am actually a bit surprised such a weak controller is paired with flash amounting to > 40-50$.
As I have several cards with manfid 0x27 and phison controllers, I started to wonder which company is actually behind this 0x27 id?
The OEM ID 0x5048 in ASCII is "PH", and I also concluded that this was Phison. Having a 64MB preferred erase size will inevitably suck for random writes, though I suspect the actual size might be smaller if you are lucky enough.
The trade-off here is to get a good linear throughput, the controller has to access several channels at once, possibly interleaving between the channels, but that means it always has to erase a whole block on each channel at once too.
A slightly smarter algorithm would be able to limit the number of blocks that have to be erased by buffering the data in RAM or SLC-type flash, but the time to erase one block on one channel is the sames as eraseing one block on each channel at once, so this is hard to find out.
Arnd
flashbench-results@lists.linaro.org