Page 1 of 1

A2 app performance class SD cards

Posted: Fri Oct 12, 2018 1:07 pm
by jahboater
I see these are now readily available:-
https://www.amazon.co.uk/SanDisk-Extrem ... mance&th=1

A1 cards were 1500 IOPS random read and 500 IOPS random write
A2 cards are 4000 IOPS random read and 2000 IOPS random write

See:
https://www.sdcard.org/developers/overv ... index.html

Juts ordered two!

Re: A2 app performance class SD cards

Posted: Fri Oct 12, 2018 1:44 pm
by jdb
+1 for App-class cards. The A1 random-read performance was often 2-3x our recommended NOOBS cards, but with a price premium.
A2 looks like a welcome bump to the critical parameter - random read IOPS.

Re: A2 app performance class SD cards

Posted: Sat Oct 13, 2018 12:38 am
by jbudd
How would the performance of Sandisk Ultra Fit USB 3.1 flash drives compare as the boot drive in a Pi 3B+ with these A2 (and A1) SD cards?
https://www.amazon.co.uk/SanDisk-Ultra- ... YCV37?th=1 64GB USB £15.84
https://www.amazon.co.uk/SanDisk-Extrem ... MBLV6?th=1 64GB uSD £21.99

Re: A2 app performance class SD cards

Posted: Sat Oct 13, 2018 4:31 am
by HawaiianPi
jbudd wrote:
Sat Oct 13, 2018 12:38 am
How would the performance of Sandisk Ultra Fit USB 3.1 flash drives compare as the boot drive in a Pi 3B+ with these A2 (and A1) SD cards?
USB flash drives typically have terrible random I/O performance and low IOPS. They are not optimized for system drive performance, so an A1 or A2 micro SD card should handily outperform most USB flash drives.

If performance is your goal, then using an SSD may make more sense. They are faster and cheaper per-GB than A2 cards (and even A1 cards in sizes over 128GB). When testing with my x86 gaming laptop on USB 3.0, a 128GB Silicon Power A55 SSD clobbered my SanDisk Ultra Plus A1 card, and it costs only £2 more than the 64GB A2 card linked in the OP (to be fair you'll also need a USB-SATA adapter cable, so it will be a little more than the £2 price difference).

SanDisk Ultra+ A1 64GB micro SD card (in USB 3.0 card reader).
  • Random Read 4KiB (Q= 8,T= 8) : 9.798 MB/s [ 2392.1 IOPS]
  • Random Write 4KiB (Q= 8,T= 8) : 2.647 MB/s [ 646.2 IOPS]
  • Random Read 4KiB (Q= 1,T= 1) : 7.622 MB/s [ 1860.8 IOPS]
  • Random Write 4KiB (Q= 1,T= 1) : 2.175 MB/s [ 531.0 IOPS]

Silicon Power A55 128GB SSD (with USB 3.0 to SATA III adapter cable).
  • Random Read 4KiB (Q= 8,T= 8) : 133.828 MB/s [ 32672.9 IOPS]
  • Random Write 4KiB (Q= 8,T= 8) : 128.656 MB/s [ 31410.2 IOPS]
  • Random Read 4KiB (Q= 1,T= 1) : 19.087 MB/s [ 4659.9 IOPS]
  • Random Write 4KiB (Q= 1,T= 1) : 39.024 MB/s [ 9527.3 IOPS]
I'd like to test it on one of my Pi3 computers, but like so many things Linux a search for benchmarks yields many results, some of which require building from source, and while I'm not completely opposed to that, it would be nice to find something that's just an apt install away.

Does anyone know of a relatively easy to install and use benchmark for Raspbian that measures IOPS and random read/write (CLI is fine)?

Re: A2 app performance class SD cards

Posted: Sat Oct 13, 2018 7:14 am
by bensimmo
This was easy enough, my SanDisk A1 cards are at the bottom, this was when the Pi3B+ was new, so before any OS alterations as new turbo/heating profiles.
I did have a few more for the different sizes.
Also Samsung EVO+ types are in the A1 class level from my test. (On a Pi3 series)

Also SD 'overclock' to 100 makes a good difference. (Also running it at a proper speed for the card)
dtparm=sd_overclock=100

https://www.jeffgeerling.com/blog/2018/ ... rison-2018


(A side note, is turn serial port off on the Pi3/ZeroW off if benchmarking and sharing, it limits the SD bus if trying to run at 100Mhz,not tested if it effects the normal usage yet.
)

Re: A2 app performance class SD cards

Posted: Sat Oct 13, 2018 7:23 am
by bensimmo
This was actually after the newer turbo/heat profile.

At none GUI boot.

SANDISK Ultra *A1* 64GB
(sdsquar-064G-GZFMA)

50Hz
1st / 2nd test
hdparm 22.51/22.61 MB/s
dd 10.0/10.1 MB/s
4k read 7878/7907 kB/s
4k write 3155/3047 kB/s

---
dtparm=sd_overclock=100

100Hz
test1/test2/test3

33.79/33.60/33.56
10.4/15.2/13.7
8979/9004/9276
3026/3160/3150

Re: A2 app performance class SD cards

Posted: Sat Oct 13, 2018 7:51 am
by HiassofT
HawaiianPi wrote:
Sat Oct 13, 2018 4:31 am
Does anyone know of a relatively easy to install and use benchmark for Raspbian that measures IOPS and random read/write (CLI is fine)?
Use iozone3 from http://www.iozone.org/ - it's available in Debian stretch (from non-free) but seems to be missing in Raspbian so you have to compile it yourself (it's easy enough though, "make linux-arm").

Then feed a bunch of command line parameters in, I use these to just check for 4k IO performance in a 100MB file - these are the same parameters as used by the people in this thread: https://forum.armbian.com/topic/954-sd- ... ment=49811 (except I removed 16k/512k/.. tests)

Code: Select all

./iozone -e -I -a -s 100M -r 4k -i 0 -i 1 -i 2 -f iotest
so long,

Hias

Re: A2 app performance class SD cards

Posted: Sat Oct 13, 2018 8:22 am
by Imperf3kt
jdb wrote:
Fri Oct 12, 2018 1:44 pm
+1 for App-class cards. The A1 random-read performance was often 2-3x our recommended NOOBS cards, but with a price premium.
A2 looks like a welcome bump to the critical parameter - random read IOPS.
Odd, over here in Australia I've been buying UHS-I cards with a slightly (and I mean slightly) lesser random read speed than these new shiny A1/A2/A3 cards for the same price (obviously I now buy A1 only)

Re: A2 app performance class SD cards

Posted: Wed Oct 17, 2018 12:22 pm
by jahboater
I can confirm the A2 class cards work fine in a Pi.
jdb wrote:
Fri Oct 12, 2018 1:44 pm
A2 looks like a welcome bump to the critical parameter - random read IOPS.
Yes, 1500 IOPS up to 4000 IOPS is a big jump!

Not done any benchmarks, but I did notice that during heavy swapping the machine was still usable (just).
Normally swapping makes a Pi appear to hang.

Re: A2 app performance class SD cards

Posted: Thu Oct 18, 2018 4:49 am
by mosespi
While doing some testing of add-on boards on a half dozen RPIs a few months ago... I noticed one was booting noticeably quicker. Turns out I had unknowingly purchased an A1 class card. Good stuff.

Regards,
-Moses

Re: A2 app performance class SD cards

Posted: Thu Oct 18, 2018 10:46 am
by runboy93
https://github.com/ThomasKaiser/Knowled ... D_cards.md
At the time of this writing due to lacking A2 host support (drivers) at least SanDisk A2 rated cards are outperformed by A1 cards from the same company (even the cheap Ultra A1 is faster everywhere except sequential write performance)
Application Performance Class 1 (A1)

This performance class requires at least 1500/500 read/write IOPS (IO operations per second) with a 4k blocksize (small data chunks) and at least 10 MB/s sustained sequential write performance. No special host requirements are needed, the card simply has to exceed the performance requirements on its own.
Application Performance Class 2 (A2)

Cache and Command Queuing require host (driver) support since the host needs to activate those new features first. The cache feature on A2 rated cards makes use of volatile RAM on the card requiring the host to learn new commands to issue flushing the cache (involving the risk of data losses -- for details see especially chapter 4.17 in Physical Layer Simplified Specification 6.0)

A1 in supported better by default than A2.

Re: A2 app performance class SD cards

Posted: Thu Oct 18, 2018 3:29 pm
by jahboater
For now - be patient!