Power consumption difference A10 vs A20

Started by hoijui, May 09, 2016, 10:17:53 AM

Previous topic - Next topic

hoijui


igorpec

Quote from: hoijui on December 09, 2016, 02:40:44 PM
despite looking in many places, and writing emails to local distributors, and looking at farnell's website, i could not yet find a single trace of the ominous SDSDSQAE cards mentioned (am in germany here).
what i was able to find, is "industrial grade" SD and MicroSD cards. there are some industrial grade SD cards with sizes up to max 4GB (not enough for a server OS + data, for me), which already cost 80+Eur, and then some industrial grade panasonic card sup to 32GB for ~170Eur, for which i could build a RAID5 out of 4 1TB SATA drives. i also found 32GB kingston "industrial (temperature)" microSD cards for around 25Eur, which would be fine, but i kind of assume that this "industrial" really just refers to the temperature range, and does not imply any improved reliability (at normal temperatures)... am i right there?
if yes, then i see no other option then to go SATA, which is a real pity, because of weight, physical size, noise and power consumption.
did i miss something, or get something wrong?

A lot of industrial grade SD cards are exactly the same as normal ones, except they are more expensive. SD media is critical element by default, no matter where you will use it. Just go for Samsung EVO and give 100EUR to charity instead. There is another general problem regarding SD cards ... there are so many fakes out there that you need a luck to get genuine one in first place. Even from respected shops.

SATA. If you are thinking of RAID with something with 25 EUR tag:

than again, forget about, since it's unreliable and RAID won't help protect your data.
linux for ARM development boards
www.armbian.com

soenke

i bought them at www.active-components.eu (i am also from germany).

I phoned with some sales guy from sandisk directly and he explained me that there is a significant difference between consumer-grade sandisk products like extreme pro and those industrial-grade cards. He said that they are more reliable and more optimized for random-access than the extreme pro. That goes with a reduced max. seq. transfer rate.

igorpec

#18
QuoteI phoned with some sales guy from sandisk directly

Well, sales people surely knows how things are ;D

BTW: We did a lot of tests with different SD cards and we were looking exactly for best "random-access":
https://forum.armbian.com/index.php/topic/954-sd-card-performance

Most expensive AKA PRO cards were not doing anything better, some of them were actually worse.

Add: you are welcome to add the same tests with those "industrial grade" cards.

Our fastest measurement was 8GB eMMC via SD, which costs 30 EUR @Pollin, Germany and I haven't saw anything close to this:
                                                              random    random
              kB  reclen    write  rewrite    read    reread    read     write
          102400       4    11485    12921     7044     7060     7010    11957
          102400      16    17668    19117    17584    17579    17467    16706
          102400     512    22711    22641    22962    22967    22942    21219
          102400    1024    22800    22871    23056    23054    23041    22737
          102400   16384    22803    22852    22990    22989    22989    22764
linux for ARM development boards
www.armbian.com

soenke

#19
This is my measurement of the SDSDSQAE with 32GB in one of our test systems.

Olimex A10S, Allwinner Kernel 3.4, ext4, nojournal, about 1 year old running linux 24/7 and a lot i/o stuff like measurement data aquisition etc.

Command line used: iozone -e -I -a -s 100M -r 4k -r 16k -r 512k -r 1024k -r 16384k -i 0 -i 1 -i 2
Output is in Kbytes/sec
Time Resolution = 0.000001 seconds.
Processor cache size set to 1024 Kbytes.
Processor cache line size set to 32 bytes.
File stride size set to 17 * record size.
                                                  random  random 
KB      reclen   write rewrite     read    reread  read   write
102400       4    2697    2817     8132     8074    8031 2674
102400      16    4770    5232    14553    14581   14055 6041
102400     512   16128   17570    19280    19274   19271 13255
102400    1024   16198   17584    19297    19294   19285   15468
102400   16384   17575   17713    19429    19505   19468   17440


Still, the main reason we chose this card is reliability. We had about 10% failing extreme pro cards before we switched to this card. So far not a single card (of some 100s) failed in the last year.

igorpec

QuoteStill, the main reason we chose this card is reliability. We had about 10% failing extreme pro cards before we switched to this card. So far not a single card (of some 100s) failed in the last year.

Reliability might be accountable, but is something hard to argue about. We don't have enough data to make any conclusions. I don't like to speculate like this: extreme pro cards were made for fast video recording, for cameras. For such usage they might not fail so soon, while they do on lot's of random write which is normal in SBC world ...

Proper reliability - "burn in" test across wide range of card is too big cost for us, so we will probably never do it unless someone puts cash on the table to cover related costs. Our small test was already on the edge of the hobby project, with cards which we had around. On top of this no one wanted to destroy them on the test :)

linux for ARM development boards
www.armbian.com

soenke

I think it is more like a selection process and some changes in parameters of the cards controller. I guess they use the upper 10% of NAND chips and controllers for more reliable cards and do a more conservative parameterization. That might also be the reason why the max. seq. transfer rate is slower.

The second somewhat 20% selection they might use in enterprise SSD, then maybe 40% for consumer SSDs. After that maybe 10% for consumer sd-cards, and the crappy rest of 10% is integrated into 3rd-party-USB-Sticks ;)

Still, the most important value of iozone for us is the 4k random read and this card is one of the fastest in this. Yes, it would be nice if random write would be faster like on a emmc, but we gladly trade this for the increased reliability and 0% dead-on-arrival-rate. So if someone has his priority on a rock-stable sd-card, chose this one.

And for the cost: it has almost the same price (even 1-2€ cheaper) as the extreme pro.