Other OS

gigabit ethernet bandwitch

6 1106


I have connected my laptop to my bananapro with an direct ethernet cable (5E 5 meters).

When I want to transfer some files from my laptop to the bananapro.

on my laptop, when I do "scp bigfile bananapi:", I usually get a speed of 9-10 Mo/s slowly decreasing to 7-8.

I checked with iperf (server is bananapi and client is my laptop) :

[  4]  0.0-10.0 sec  1.06 GBytes   907 Mbits/sec

and also with ethtool, on both side I have :
    Speed: 1000Mb/s
    Duplex: Full

If I do from the bananapi "scp laptop:bigfile .", I get slightly better : 10-12 Mo/s staying constant.

I thought that such a direct Gigabit link could achieve much better bandwitch.

On the bananapi i read/write a SD card which can theoretically reach 90-95 Mo/s

but when I run  "hdparm -t /dev/mmcblk0"

I get

Timing buffered disk reads:  54 MB in  3.10 seconds =  17.43 MB/sec

is the network bandiwtch limited by the IO on the bananapi ?

What can I do to improve the bandwitch ?

also, I noticed that when I run previous scp commands, one of the bananapi cpu hits 100%, could it also be a limitation ?

Thanks for your help.

Best regards

Ps : armbian jessie

You just find out hardware limitations ;( One is cheap SD card controller, than you have CPU, sata r/w, ...

Edited by deb75 at Mar 10, 2016 02:44


Yes, I believe the micro SD card IO bandwitch is mainly responsible for the network limitation,
but I am very surprised as I purchased a quality (supposed to be) micro SD card : MicroSDHC SanDisk Extreme Plus 32 Go up to 95 Mo/s, Classe 10, U3

So my first issue to solve is understand why I get so poor IO with this micro SD card, perhaps using "hdparm -t /dev/mmcblkp0" is not the appropriate tool (?)

Another point is the CPU : I should try another transfer application than ssh/sshd for big files to avoid the CPU crypting/decrypting burden, or perhaps this one could get accelerated with the sunxi-ss module ?

My main concern is the micro SD card, I just cannot accept such poor reading/writing for what I paid.

Any suggestions are welcomed.

Best regards.

EDIT : I am all the more surprised of the micro SD card performance as the boot process is really fast (a few seconds), in accordance with what I expected from this micro SD card.


I have just read this :
SD Card: limited to max. 16.x MB/s sequential speed on A10/A20, random I/O differs heavily and is only dependant on the SD card in use

from the page you indicated. So I guess there is no hope for SD cards. I will try to read/write from the SATA hdd drive.

Edited by igorpec at Mar 10, 2016 13:06

Limitations are everywhere. You can read about on that page or find out by yourself - A20 boards have actually quite good SATA utilization (40 write, 100 read), ethernet is close to gigabit, so don't rush to buy yet another useless board. Try the same on Raspberry PI or any other non A20 Allwinner board and you will cry A20 is actually not that bad.

Regarding crypting/decrypting with sunxi-ss ... don't have experiences.

If you want proper performances, check Marvell based boards: Clearfog, Turris Omnia. Their utilization goes to maximum everywhere and you can boot from M2 drive with 500 + MB/s without any problems ;)

Edited by Tido at Mar 10, 2016 21:04
deb75 replied at Mar 10, 2016 10:08
MicroSDHC SanDisk Extreme Plus 32 Go up to 95 Mo/s, Classe 10, U3


Maybe the first sentence in this posting is interesting for you.

What do you mean with: Mo/s ??



I finally reached better network performances.

The micro SD card is replaced by a sata drive correctly set with hdparm.
The ssh protocol is forgotten in favour of a more direct transfer approach :
- on the bananapi : nc -l ipadress port | pv -pterba -s size_of_file > output_file
- on the laptop : nc ipadress port < input_file

Like this, I get an average bandwitch around 20 Mo/s (1 Mo = 1 MB = 1 Mega Bytes = 8 Mb = 8 Megabits.

I guess it could be further improved by applying the settings http://forum.lemaker.org/thread-7102-1-1.html

You have to log in before you can reply Login | Sign Up

Points Rules