I'll try to address your first question regarding the storage device durability as I'm a bit familiar with that.
Switching from SD to eMMC might not improve the situation if you don't do an accessment of your system's storage usage and take action to improve things, because both SD and eMMC use NAND.
Do you have an estimate of data writes to your storage?
Use the following to evaluate your use case [see [1] for details]
total bytes written throughout device life = (device capacity in bytes) * (max program/erase cycles) / (write amplification factor)
Say for example
- you write 0.5GiB per day
- want your device to operate for 5 years
- partitions you write data to totals 4GiB (storage capacity is more
than this, but other partitions are read-only)
- max program/erase cycles is 3000 for your multi-level cell (MLC) NAND
This gives you a write amplification factor of
4 * 3000 / (0.5 * 365 * 5) = ~13
What is write amplification
NAND in the SD or eMMC is written in NAND pages. Suppose you write/modify 1KiB (two 512-byte sectors) from the host, but say NAND page is 16KiB. So, the eMMC controller will write a whole NAND page.
Things get more complicated when you think of erasures, because NAND is erased in NAND blocks, and a NAND block consists of many NAND pages.
So, what can you do to improve device life
From the above equation, you can
- increase device capacity (but that'll add to the cost)
- improve program/erase cycles: go for SLC or turn your data write partitions from MLC to pSLC (but this reduces the capacity)
- reduce write amplification by improving your apps to perform NAND page aligned, NAND page sized (or multiples) writes from host (see eMMC EXT_CSD[265] optimal write size), enabling eMMC cache etc.
What else can you do
You can monitor your eMMC health using mmc-utils (https://git.kernel.org/pub/scm/utils/mmc/mmc-utils.git) or sysfs, and take necessary steps before the failure comes as a surprise.
- eMMC extended CSD register provides
- estimate for life time of SLC and MLC blocks in steps of 10%
(0x01 = 0-10% device life time used, 0x02 = 10-20%, .. , 0x0B = end of life)
type-B (MLC): EXT_CSD[268], type-A (SLC): EXT_CSD[269]
- status of the spare blocks that are used to replace bad blocks
(0x01: Normal, 0x02: Warning: 80% of blocks used, 0x03: Urgent: 90% of blocks used)
Pre EOL info: EXT_CSD[267]
- vendor may provide a proprietary health report in EXT_CSD[301:270] (but so far, I have only seen all zeros here)
e.g.
mmc-utils:
# mmc extcsd read /dev/mmcblk0
:
eMMC Life Time Estimation A [EXT_CSD_DEVICE_LIFE_TIME_EST_TYP_A]: 0x01
eMMC Life Time Estimation B [EXT_CSD_DEVICE_LIFE_TIME_EST_TYP_B]: 0x00
eMMC Pre EOL information [EXT_CSD_PRE_EOL_INFO]: 0x01
:
sysfs:
# cat /sys/block/mmcblk0/device/life_time
0x00 0x01
# cat /sys/block/mmcblk0/device/pre_eol_info
0x01
- vendor may provide health related information you can access from mmc generic command CMD56 (using mmc-utils, mmc gen_cmd read < device > [arg])
See the following for a good explanation:
[1] https://www.kingston.com/en/embedded/emmc-embedded-flash
'Estimating, validating & monitoring eMMC life cycle'
