Evo Plus 64GB - Poor Random IO

Dan25 · November 7, 2016

Hey, just wondering if this is a common issue, I've put a fresh copy of Armbian 5.20 (legacy kernel) on my new Evo Plus 64GB and updated it to 5.23 on my Orange Pi PC, but it seems I'm getting only 742KB/s 4k random write speeds (I thought it'd be around 2MB/s) and 5 to 10MB/s sequential write speed.

Is this normal, and if so is there a better card to buy that'll give me more Random IO & saturate the write speeds of the bus on the Opi PC?

The results from testing /dev/mmcblk0p1 (ext4):
  Data OK: 55.34 GB (116061392 sectors)
Data LOST: 0.00 Byte (0 sectors)
Average writing speed: 10.05 MB/s
Average reading speed: 20.35 MB/s
                                            random    random
reclen    write  rewrite    read    reread    read     write
     4      690      687     4498     4203     4158      742                                                          
   512     2404     6939    21455    21446    21299     8556                                                          
 16384    10515    12075    22813    22800    22662    10040                                                          

Health summary: OK

Performance summary:
Sequential reading speed: 20.35 MB/s 
 4K random reading speed:  4158 KB/s 
Sequential writing speed: 10.05 MB/s 
 4K random writing speed:   742 KB/s (too low)

hojnikb · November 7, 2016

What test did you use ?

where did you buy the card ?

tkaiser · November 7, 2016

Hmm... could you please repeat the test, this time running in one terminal line 1 and in another line 2:

cd $HOME ; iozone -e -I -a -s 100M -r 4k -r 16k -r 512k -r 1024k -r 16384k -i 0 -i 1 -i 2
sudo armbianmonitor -m

The armbianmonitor call monitors both cpufreq and should also display %iowait percentage. And cpufreq should remain at 1296 MHz throughout the whole test. Please report back with results.

Dan25 · November 8, 2016

On 11/7/2016 at 4:22 PM, hojnikb said:

What test did you use ?

where did you buy the card ?

I used armbianmonitor -c, and the card was bought from an Amazon seller and fulfilled by Amazon.

On 11/7/2016 at 4:32 PM, tkaiser said:
Hmm... could you please repeat the test, this time running in one terminal line 1 and in another line 2:

Sure, here are my results:

Reveal hidden contents

And CPU stats here: https://ncry.pt/p/xNCn#WmGhevkPANlkni8RZwupGCuoVUaBIQi7-xvQNd5imek

tkaiser · November 8, 2016

These numbers are clearly bad and do not match every other EVO result from 2016. Would be interesting how mmc info looks like ('sudo armbianmonitor -u')

Dan25 · November 8, 2016

Sure, definitely: http://sprunge.us/cABh

I ordered a few Evo 32GB cards for my other OrangePI PC's to see if they'll have better random IO than this card. Will definitely test them before using them though to make sure they are the labeled capacity and perform properly.

tkaiser · November 8, 2016

On 11/8/2016 at 10:41 AM, Dan25 said:

Sure, definitely: http://sprunge.us/cABh

preferred_erase_size looks suspicious to me (4 times more on all EVO/EVO+ so far). I would test the card also with a fast device and if sequential speeds do not exceed what's guaranteed by specs immediately return it mentioning mmc info and benchmark results.

Dan25 · November 8, 2016

I asked Amazon yesterday about the poor IO performance, they refunded me immediately after telling me that the whole category of PC Accessories doesn't qualify for returns. Told them they were 2 for 4 on bad hardware shipped to me in the last 6 months, and that I'd not be shopping at their new Amazon Fresh store they're opening here in Seattle due to the risk of their supply chain giving me E. coli covered veggies.

Should I just plug it into my laptop (i5 w/SSD) via a USB 2.0 dongle and cd into /media/dan/<insert UUID here> and run the same IO test below?

iozone -e -I -a -s 100M -r 4k -r 16k -r 512k -r 1024k -r 16384k -i 0 -i 1 -i 2

Edit: So I ran a benchmark from my laptop, seems better in sequential and worse in random IO. Perhaps I'm missing something but I can't find preferred_erase_size, and the microSD is showing up under /sys/block/sda/ and not as /sys/block/mmcblk0/ (my SSD is mounted on sdb). Perhaps its this janky AF 5 in 1 card reader from Frys with an integrated flexy USB cable that is making it mount somewhat oddly?

Reveal hidden contents

tkaiser · November 8, 2016

Well, it would be good to know whether the card reader is a bottleneck or not. My laptop (MacBook Pro) is able to write/read with +80 MB/s but the only USB card reader I have starts to throttle and/or corrupt data after some time when used with USB3 ports. But you could try it anyway (though not that important since you already got a refund).

When your 32 GB EVOs arrive could you please check preferred_erase_size and report back?

Dan25 · November 8, 2016

Will do, where'd preferred_erase_size be when attached via USB on a standard Jessie install? Gonna dig around for another mSD reader here and give that a go.

tkaiser · November 8, 2016

Sorry, have been wrong here anyway, see link (32 GB EVO used booting OPi Zero): http://forum.armbian.com/index.php/topic/2808-orange-pi-zero-went-to-the-market/page-2#entry19293

Dan25 · November 8, 2016

Huh? I see the mSD card looks like an Evo on post #29 there, but I don't see much other pertinent info...

I found 2 other mSD adapters, and a 4 pack of Sandisk mSD cards, benchmarks below. Still don't see any file called preferred_erase_size in /sys, and /sys/block just has sda, sdb & sdc.

Evo Plus 64GB:

Reveal hidden contents

Sandisk 16GB:

Reveal hidden contents

dan@dax:/media/dan/6331-6630$ sudo iozone -e -I -a -s 100M -r 4k -r 16k -r 512k -r 1024k -r 16384k -i 0 -i 1 -i 2
[sudo] password for dan:
   Iozone: Performance Test of File I/O
            Version $Revision: 3.429 $
       Compiled for 64 bit mode.
       Build: linux-AMD64

   Contributors:William Norcott, Don Capps, Isom Crawford, Kirby Collins
                 Al Slater, Scott Rhine, Mike Wisner, Ken Goss
                 Steve Landherr, Brad Smith, Mark Kelly, Dr. Alain CYR,
                 Randy Dunlap, Mark Montague, Dan Million, Gavin Brebner,
                 Jean-Marc Zucconi, Jeff Blomberg, Benny Halevy, Dave Boone,
                 Erik Habbinga, Kris Strecker, Walter Wong, Joshua Root,
                 Fabrice Bacchella, Zhenghua Xue, Qin Li, Darren Sawyer,
                 Vangel Bojaxhi, Ben England, Vikentsi Lapa.

   Run began: Tue Nov 8 05:53:43 2016

   Include fsync in write timing
   O_DIRECT feature enabled
   Auto Mode
   File size set to 102400 kB
   Record Size 4 kB
   Record Size 16 kB
   Record Size 512 kB
   Record Size 1024 kB
   Record Size 16384 kB
   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     bkwd    record    stride
              kB reclen    write rewrite    read    reread    read     write     read   rewrite      read   fwrite frewrite    fread freread
          102400       4     3258     2955     4621     4670     4599     1394
          102400      16     7345     7583    11172    11194    11176     1519
          102400     512    11302    10935    18943    18903    18986     2584
          102400    1024    11124    13174    19405    19462    19326     5025
          102400   16384    11693    13015    19560    19489    19551    12140

iozone test complete.

Dan25 · November 11, 2016

So I got 5 of these used Evo 32GB MicroSD cards on eBay, and while 3 of them are hitting 3MB/s in 4k random IO, the other two are pegging in around 800KB/s. Commonalities I noticed are the good ones are made in Korea (part MBMPBGVEDACW-F, 1 tested) & Philippines (part MBMPBGVEODFW-F, 2 tested), and the slow Evo cards were made in the Philippines (part MBMPBGVEDFCW-P, 2 tested). The slow ones passed f3probe, so I'm wondering if its just age/controller failure causing the very poor 4k random IO performance.

Reveal hidden contents

dan@dax:/media/dan/F295-DC34$ iozone -e -I -a -s 100M -r 4k -r 16k -r 512k -r 1024k -r 16384k -i 0 -i 1 -i 2
   Iozone: Performance Test of File I/O
            Version $Revision: 3.429 $
       Compiled for 64 bit mode.
       Build: linux-AMD64

   Contributors:William Norcott, Don Capps, Isom Crawford, Kirby Collins
                 Al Slater, Scott Rhine, Mike Wisner, Ken Goss
                 Steve Landherr, Brad Smith, Mark Kelly, Dr. Alain CYR,
                 Randy Dunlap, Mark Montague, Dan Million, Gavin Brebner,
                 Jean-Marc Zucconi, Jeff Blomberg, Benny Halevy, Dave Boone,
                 Erik Habbinga, Kris Strecker, Walter Wong, Joshua Root,
                 Fabrice Bacchella, Zhenghua Xue, Qin Li, Darren Sawyer,
                 Vangel Bojaxhi, Ben England, Vikentsi Lapa.

   Run began: Fri Nov 11 02:38:00 2016

   Include fsync in write timing
   O_DIRECT feature enabled
   Auto Mode
   File size set to 102400 kB
   Record Size 4 kB
   Record Size 16 kB
   Record Size 512 kB
   Record Size 1024 kB
   Record Size 16384 kB
   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     bkwd    record    stride
              kB reclen    write rewrite    read    reread    read     write     read   rewrite      read   fwrite frewrite    fread freread
          102400       4     1178     1118     3066     3074     3041      829
          102400      16     2646     3104     9306     9319     9274     2601
          102400     512     7582     9476    18293    18233    18235     5254
          102400    1024     6842     7032    18648    18667    18496     5610
          102400   16384     6337     7490    18615    18582    18552     7425

iozone test complete.
dan@dax:/media/dan/F295-DC34$ cd ..
dan@dax:/media/dan$ cd E4DB-2005/
dan@dax:/media/dan/E4DB-2005$ iozone -e -I -a -s 100M -r 4k -r 16k -r 512k -r 1024k -r 16384k -i 0 -i 1 -i 2
   Iozone: Performance Test of File I/O
            Version $Revision: 3.429 $
       Compiled for 64 bit mode.
       Build: linux-AMD64

   Contributors:William Norcott, Don Capps, Isom Crawford, Kirby Collins
                 Al Slater, Scott Rhine, Mike Wisner, Ken Goss
                 Steve Landherr, Brad Smith, Mark Kelly, Dr. Alain CYR,
                 Randy Dunlap, Mark Montague, Dan Million, Gavin Brebner,
                 Jean-Marc Zucconi, Jeff Blomberg, Benny Halevy, Dave Boone,
                 Erik Habbinga, Kris Strecker, Walter Wong, Joshua Root,
                 Fabrice Bacchella, Zhenghua Xue, Qin Li, Darren Sawyer,
                 Vangel Bojaxhi, Ben England, Vikentsi Lapa.

   Run began: Fri Nov 11 02:53:59 2016

   Include fsync in write timing
   O_DIRECT feature enabled
   Auto Mode
   File size set to 102400 kB
   Record Size 4 kB
   Record Size 16 kB
   Record Size 512 kB
   Record Size 1024 kB
   Record Size 16384 kB
   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     bkwd    record    stride
              kB reclen    write rewrite    read    reread    read     write     read   rewrite      read   fwrite frewrite    fread freread
          102400       4     2989     2872     4616     4616     4688     3061
          102400      16     6876     7035    11455    11422    11300     6619
          102400     512    13601    15052    19611    19555    19566    13941
          102400    1024    15183    14987    19894    19837    19799    14193
          102400   16384    14771    14415    19886    20003    19924    16222

iozone test complete.
dan@dax:/media/dan/E4DB-2005$ cd
dan@dax:~$ cd /media/dan/C330-9944/
dan@dax:/media/dan/C330-9944$ iozone -e -I -a -s 100M -r 4k -r 16k -r 512k -r 1024k -r 16384k -i 0 -i 1 -i 2
   Iozone: Performance Test of File I/O
            Version $Revision: 3.429 $
       Compiled for 64 bit mode.
       Build: linux-AMD64

   Contributors:William Norcott, Don Capps, Isom Crawford, Kirby Collins
                 Al Slater, Scott Rhine, Mike Wisner, Ken Goss
                 Steve Landherr, Brad Smith, Mark Kelly, Dr. Alain CYR,
                 Randy Dunlap, Mark Montague, Dan Million, Gavin Brebner,
                 Jean-Marc Zucconi, Jeff Blomberg, Benny Halevy, Dave Boone,
                 Erik Habbinga, Kris Strecker, Walter Wong, Joshua Root,
                 Fabrice Bacchella, Zhenghua Xue, Qin Li, Darren Sawyer,
                 Vangel Bojaxhi, Ben England, Vikentsi Lapa.

   Run began: Fri Nov 11 03:01:24 2016

   Include fsync in write timing
   O_DIRECT feature enabled
   Auto Mode
   File size set to 102400 kB
   Record Size 4 kB
   Record Size 16 kB
   Record Size 512 kB
   Record Size 1024 kB
   Record Size 16384 kB
   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     bkwd    record    stride
              kB reclen    write rewrite    read    reread    read     write     read   rewrite      read   fwrite frewrite    fread freread
          102400       4     3246     3093     4622     4649     4634     3198
          102400      16     9659    10073    11820    11804    11734    11585
          102400     512    13283    16536    19202    19354    19249    16880
          102400    1024    13084    16275    19774    19786    19751    17199
          102400   16384    13803    16599    19698    19769    19719    17202

iozone test complete.
dan@dax:/media/dan/C330-9944$ cd ..
dan@dax:/media/dan$ cd A1D0-A2BC/`
> ^C
dan@dax:/media/dan$ cd A1D0-A2BC/
dan@dax:/media/dan/A1D0-A2BC$ iozone -e -I -a -s 100M -r 4k -r 16k -r 512k -r 1024k -r 16384k -i 0 -i 1 -i 2
   Iozone: Performance Test of File I/O
            Version $Revision: 3.429 $
       Compiled for 64 bit mode.
       Build: linux-AMD64

   Contributors:William Norcott, Don Capps, Isom Crawford, Kirby Collins
                 Al Slater, Scott Rhine, Mike Wisner, Ken Goss
                 Steve Landherr, Brad Smith, Mark Kelly, Dr. Alain CYR,
                 Randy Dunlap, Mark Montague, Dan Million, Gavin Brebner,
                 Jean-Marc Zucconi, Jeff Blomberg, Benny Halevy, Dave Boone,
                 Erik Habbinga, Kris Strecker, Walter Wong, Joshua Root,
                 Fabrice Bacchella, Zhenghua Xue, Qin Li, Darren Sawyer,
                 Vangel Bojaxhi, Ben England, Vikentsi Lapa.

   Run began: Fri Nov 11 03:09:58 2016

   Include fsync in write timing
   O_DIRECT feature enabled
   Auto Mode
   File size set to 102400 kB
   Record Size 4 kB
   Record Size 16 kB
   Record Size 512 kB
   Record Size 1024 kB
   Record Size 16384 kB
   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     bkwd    record    stride
              kB reclen    write rewrite    read    reread    read     write     read   rewrite      read   fwrite frewrite    fread freread
          102400       4     1261     1284     3138     3118     3112      879
          102400      16     4105     3747     9517     9437     9226     2988
          102400     512     8224     9875    18186    18182    18157     5096
          102400    1024     6585     8227    18711    18681    18575     5670
          102400   16384     6427     7372    18647    18652    18559     6298

iozone test complete.
dan@dax:/media/dan/A1D0-A2BC$ cd ..
dan@dax:/media/dan$ cd
89AF-51A0/
dan@dax:/media/dan$ cd 89AF-51A0/
dan@dax:/media/dan/89AF-51A0$ iozone -e -I -a -s 100M -r 4k -r 16k -r 512k -r 1024k -r 16384k -i 0 -i 1 -i 2
   Iozone: Performance Test of File I/O
            Version $Revision: 3.429 $
       Compiled for 64 bit mode.
       Build: linux-AMD64

   Contributors:William Norcott, Don Capps, Isom Crawford, Kirby Collins
                 Al Slater, Scott Rhine, Mike Wisner, Ken Goss
                 Steve Landherr, Brad Smith, Mark Kelly, Dr. Alain CYR,
                 Randy Dunlap, Mark Montague, Dan Million, Gavin Brebner,
                 Jean-Marc Zucconi, Jeff Blomberg, Benny Halevy, Dave Boone,
                 Erik Habbinga, Kris Strecker, Walter Wong, Joshua Root,
                 Fabrice Bacchella, Zhenghua Xue, Qin Li, Darren Sawyer,
                 Vangel Bojaxhi, Ben England, Vikentsi Lapa.

   Run began: Fri Nov 11 03:22:58 2016

   Include fsync in write timing
   O_DIRECT feature enabled
   Auto Mode
   File size set to 102400 kB
   Record Size 4 kB
   Record Size 16 kB
   Record Size 512 kB
   Record Size 1024 kB
   Record Size 16384 kB
   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     bkwd    record    stride
              kB reclen    write rewrite    read    reread    read     write     read   rewrite      read   fwrite frewrite    fread freread
          102400       4     3168     3087     4639     4650     4600     3193
          102400      16     9747    10092    11438    11360    11411    11824
          102400     512    11770    17038    19529    19614    19580    16946
          102400    1024    13076    16323    19769    19794    19634    17186
          102400   16384    13749    16595    19827    19825    19811    17127

iozone test complete.

f3probe results for the 2 slow cards:

Reveal hidden contents

hojnikb · November 11, 2016

Maybe it's a different controller alltogether or tuned for sequential performance rather than random.

Dan25 · November 11, 2016

Eh, I'd hope it'd perform better, cause the slow cards seem to top out at 6MB/s therefore they don't meet the class 10 rating stamped on the card.

Sign In

Evo Plus 64GB - Poor Random IO

Recommended Posts

Dan25

hojnikb

tkaiser

Dan25

tkaiser

Dan25

tkaiser

Dan25

tkaiser

Dan25

tkaiser

Dan25

Dan25

hojnikb

Dan25

Forums

My Activity Streams

Download

Store

Important Information