USB-to-SATA performance on OPi Plus

bronco · Published in 2015-12-1 05:04:13

Edited by bronco at 2015-12-1 15:11

Can you please share your performance results with a connected SATA disk attached to the USB-to-SATA bridge on the Plus?

I'm currently reviewing the new Banana Pi M3 which uses the same GL830 bridge and it's a bit hard to believe how slow disk accesses are: I do not exceed 13 MB/s write and 23 MB/s read using iozone.

Would be great if you could provide also some performance numbers from the OPi plus to confirm that. Please no dd/hdparm nonsense but iozone instead (apt-get install iozone3). The first 2 iozone calls from here http://linux-sunxi.org/USB/UAS#Testing_3_different_external_enclosures would be interesting to compare with.

Thx

miha · Published in 2015-12-1 15:35:32

Hi,

Installed izone to my CubieTruck to get some values from there for the reference. My cubietruck is running ubuntu 14.04 LTS "server-version" ie. no X-window and OS runs on SSD.

      Using maximum file size of 4096000 kilobytes.
      File size set to 4096000 kB
      Record Size 4 kB
      Command line used: ./iozone -a -g 4000m -s 4000m -i 0 -i 1 -r 4K
      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
      4096000    4 39513 40234 133816 144580
#########

      Using maximum file size of 4096000 kilobytes.
      File size set to 4096000 kB
      Record Size 1024 kB
      Command line used: ./iozone -a -g 4000m -s 4000m -i 0 -i 1 -r 1024K
      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
      4096000 1024 39536 40186 118838 117250
#########

      Command line used: ./iozone -O -i 0 -i 1 -i 2 -e -+n -r 4K -s 2000m
      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
      2048000    4    9747       0 34491       0    4420    2620

CPU details:
# cat /proc/cpuinfo
Processor       : ARMv7 Processor rev 4 (v7l)
processor       : 0
BogoMIPS       : 1816.98

processor       : 1
BogoMIPS       : 1823.53

Features       : swp half thumb fastmult vfp edsp neon vfpv3 tls vfpv4 idiva idivt
CPU implementer       : 0x41
CPU architecture: 7
CPU variant       : 0x0
CPU part       : 0xc07
CPU revision       : 4

Hardware       : sun7i
Revision       : 0000
Serial             : 0000000000000000

Br,
-miha

miha · Published in 2015-12-1 15:52:19

Oh, forgot to mention that I don't own Orange yet but one piece ordered from aliexpress is on the way "015.11.29 17:27 (GMT-7): In transit (shipped by air)".
If parcel does not stop to customs I'll receive my Orange Pi Plus quite soon...

After I have managed to install OP+ and iozone, I'll post my results from OP+ also.
Thanks.

bronco · Published in 2015-12-1 16:08:23

miha replied at 2015-12-1 15:35
Installed izone to my CubieTruck to get some values from there for the reference. My cubietruck ...

Thx for the reference values. But you should add that your A20 based Cubietruck has a real SATA port and not just an USB-to-SATA bridge as the H3 based OPi +.

BTW: Using mainline kernel and A20 based boards with fast DRAM clock you'll be able to get close to 45/200 MB/s through SATA.

Really looking forward to your iozone results from OPi +. I've to admit that I retested yesterday again with the BPi-M3 to get a clue how cpufreq settings matter (must also apply to H3): Two times sequential transfer speeds, 1st @ 480 MHz, 2nd @ 1800 MHz:

480 MHz: kB reclen write rewrite read reread
4096000 4 13529 13466 22393 22516
4096000 1024 13588 13411 22717 26115
1.8 GHz: 4096000 4 15090 15082 30968 30316
4096000 1024 15174 15131 30858 29441

Copy code

fritz · Published in 2015-12-2 17:52:37

Well
with mainline 4.3 kernel :

   KB  reclen write  rewrite    read reread
      4096000    4 15123 15084 31005 31042

               KB  reclen    write  rewrite    read reread
      4096000 1024 15133 15237 30504 30514

Then with the only debian based distro i had on my Harddisk (lubuntu 14.04 untouched):
With 3.4 and scaling_governor at interactive:

   KB  reclen write  rewrite    read reread
      4096000    4 15383 15588 31442 31402

               KB  reclen    write  rewrite    read reread
      4096000 1024 15445 15260 20417 31516

With 3.4 and scaling_governor at performance (1200):

                  KB  reclen write  rewrite    read reread
      4096000    4 13753 13399 31252 31276

And now The Big Bang with a simple USB 2.0 2.5"Harddrive attached at the USB-port:

KB reclen write rewrite read reread
4096000 4 36752 36004 39533 39573

As I said this USB-SATA bridge is slower than a "simple" USB-Harddisk.

bronco · Published in 2015-12-2 18:24:55

fritz replied at 2015-12-2 17:52
And now The Big Bang with a simple USB 2.0 2.5"Harddrive attached at the USB-port:

Thx for both confirmation of the bad GL830 performance and that USB support in mainline kernel for M3 is ready (the good news: you'll be able to get close to 40MB/s using an UAS capable USB-to-SATA bridge... and of course avoiding the GL830 at all).

I made the same experience with Banana Pi M3 when comparing onboard 'pseudo SATA' with a random external enclosure: Close to 35MB/s: http://forum.armbian.com/index.p ... ew-of-banana-pi-m3/

It seems, the GL830 is a really bad choice to provide 'pseudo SATA' on OPi + and Banana Pi M3.

miha · Published in 2015-12-12 04:07:41

Orange Pi Plus received. Wheezy installed to EMMC and 120GB SSD hooked.

And this is result:

Run began: Fri Dec 11 19:52:58 2015

Auto Mode
Using maximum file size of 4096000 kilobytes.
File size set to 4096000 KB
Record Size 4 KB
Command line used: iozone -a -g 4000m -s 4000m -i 0 -i 1 -r 4K
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
      4096000    4
Error writing block 398070, fd= 3
write: Read-only file system

iozone: interrupted

exiting iozone

root@OrangePI:/mnt/home# dmesg
block 408792
[ 2226.210094] Buffer I/O error on device sda1, logical block 408793
[ 2226.210094] Buffer I/O error on device sda1, logical block 408794
[ 2226.210094] Buffer I/O error on device sda1, logical block 408795
[ 2226.210094] Buffer I/O error on device sda1, logical block 408796
[ 2226.210094] Buffer I/O error on device sda1, logical block 408797
[ 2226.210094] Buffer I/O error on device sda1, logical block 408798
[ 2226.210094] Buffer I/O error on device sda1, logical block 408799
---clip---
[ 2226.239402] Aborting journal on device sda1-8.
[ 2226.239431] JBD2: Error -5 detected when updating journal superblock for sda1-8.
[ 2226.241105] EXT4-fs error (device sda1): ext4_journal_start_sb:328: Detected aborted journal
[ 2226.241120] EXT4-fs (sda1): Remounting filesystem read-only
[ 2226.241128] EXT4-fs (sda1): previous I/O error to superblock detected
[ 2226.241178] EXT4-fs (sda1): ext4_da_writepages: jbd2_start: 2147483647 pages, ino 12; err -30
[ 2226.270476] EXT4-fs error (device sda1): ext4_find_entry:935: inode #2: comm iozone: reading directory block 0

Linux OrangePI 3.4.39 #2 SMP PREEMPT Mon Oct 12 12:03:03 CEST 2015 armv7l GNU/Linux

No heatsink (not received yet) and power supply provides 2.1A. Need maybe more juice?

I have regulated 5V PSU which can provide 6A but needs some soldering, replacing barrel plug with correct size for the OPI+

Br
-miha

miha · Published in 2015-12-12 04:15:04

Hi,

SSD disappeared completely from system. fdisk did not show device at all. partprobe didn't help but after reboot SSD is back. Stopped testing with iozone, trying again after PSU replaced.

Br
-miha

miha · Published in 2015-12-12 17:24:26

Edited by miha at 2015-12-12 18:51

Replacing PSU with more powerful did the trick, no more I/O errors. I did temporary soldering to get fast confirmation for my assumption that 2.1A is not enough for the OPI+ w/ SSD.

Still running without heatsink and these commands issued before running iozone:

# echo 1008000 > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
# echo 1008000 > /sys/devices/system/cpu/cpu1/cpufreq/scaling_max_freq
# echo 1008000 > /sys/devices/system/cpu/cpu2/cpufreq/scaling_max_freq
# echo 1008000 > /sys/devices/system/cpu/cpu3/cpufreq/scaling_max_freq

Copy code

And the results from iozone:

Command line used: iozone -a -g 4000m -s 4000m -i 0 -i 1 -r 4K
random random bkwd record stride
KB reclen write rewrite read reread read write read rewrite read fwrite frewrite fread freread
4096000 4 14746 14816 32717 32792

Copy code

Command line used: iozone -a -g 4000m -s 4000m -i 0 -i 1 -r 1024K
.
random random bkwd record stride
KB reclen write rewrite read reread read write read rewrite read fwrite frewrite fread freread
4096000 1024 15525 15574 34425 34658

Copy code

Command line used: iozone -O -i 0 -i 1 -i 2 -e -+n -r 4K -s 2000m
random random bkwd record stride
KB reclen write rewrite read reread read write read rewrite read fwrite frewrite read reread
2048000 4 3722 0 10079 0 1419 1674

Copy code

SSD is mounted with options discard,nodiratime,noatime. Filesystem is ext4

bronco · Published in 2015-12-12 19:51:08

miha replied at 2015-12-12 17:24

I did temporary soldering to get fast confirmation for my assumption that 2.1A is not enough for the OPI+ w/ SSD.

2.1A should be enough even if you use the wrong dvfs settings (all OS images currently use). Even when stresstesting an SSD and running at the upper clockspeed limits while being overvolted you should not exceed 8W (1.6A).

How did you supply power before?