Question 1: There are three multiplexed ADC channels in the Airspy, is there a simple way to access the other two if I wanted to implement a custom DDC firmware?
Answer 1: .By default channel 0 is used for the R820T2, but channel 1 and channel 2
are unconnected. You could hack the firmware source code and modify
this line "#define DEFAULT_ADCHS_CHAN (0)" in
"firmware/airspy_m4/airspy_m4.c" to use channel 1 or channel 2 instead
of channel 0 which is connected to the output of the R820T2 tuner chip.
Of course the R820T2 would still be powered on so you should disable it,
since you are not using it (use disable_r820t_power() anywhere where
enable_r820t_power() is called) because it may add unwanted noise. Of course disabling the R820T2 chip, also means that you should
delete/disable any calls to set registers in it, since you are not be
using it. You should also delete/disable the frequency tuning from the firmware since the tuner is not used. And by the time you have done that it might be easier just to
write your own custom firmware.
Extra info 1: Since the spyverter is now available it might be easier just to use it unless you actually need an ADC, that in theory could run at 80MSPS. The problem is that the USB bus will limit the data transfer to about 20MSPS. And there is not enough RAM in the LPC4370 to buffer more than about 2x16KiB (double buffering) or 32KiB of data, and there is not enough processing power within the CPU to reduce this data. So unless 32KiB of data sampled at 80MSPS is enough for your function this is probably a dead end for your idea. If you are still going to use the two ADC ports the good news is that there is ESD protection in the R2 board, the R0 board had no ESD protection on the ADC inputs.
Sunday, 29 November 2015
Monday, 2 March 2015
How to use Kal software to workout the PPM frequency offset of a RTL-SDR
The basic idea behind the Kalibrate software is that a GSM mobile phone base station has an expensive very high precision atomic clock (0.01ppb), where as a RTL-SDR stick has a cheap low precision 28.8MHz quartz crystal (+/-100ppm). This cheap low precision oscillator is used to derive all frequencies when you tune the RTL-SDR, so the bigger the errors the more you are offset from the correct frequency. There are two pieces of good news the offset is linear and once the cheap crystal has warmed up it generally does not drift very much. Because the error is linear across the whole frequency range, only one correction value is required.
When you buy a 28.8MHz quartz crystal in a can, they are generally not 28,800,000Hz. There are two problems that need to be compensated for by the PPM offset in SDR receiver software. The first is manufacturing tolerance, generally about 100 PPM (0.01%) is the range to keep costs down, so the frequency of a new quartz crystal could be anywhere between 28,797,120Hz and 28,802,880Hz at 20 degrees Celsius. The second problem is that the frequency will change slightly as the device warms up. But the good news is once the device has warmed up it will not drift much, unless the ambient room temperature varies.
So basically you scan the local area for active GSM base station channels with a large amount of power (good signal strength, which will probably be from the base station that is physically closest to your current location). Then kal uses this channels GSM frequency correction bursts to workout the PPM (part per million) offset of the RTL-SDR that you are currently calibrating.
$ git clone https://github.com/steve-m/kalibrate-rtl
$ cd kalibrate-rtl
$ ./bootstrap && CXXFLAGS='-W -Wall -O3'
$ ./configure
$ make
$ sudo make install
(confirm that the RTL-SDR is actually plugged into the Linux machine)
$ lsusb | grep -i RTL
Bus 004 Device 015: ID 0bda:2838 Realtek Semiconductor Corp. RTL2838 DVB-T
(print the help for the kal software to see what arguments should be used)
$ kal
error: must enter channel or frequency
kalibrate v0.4.1-rtl, Copyright (c) 2010, Joshua Lackey
modified for use with rtl-sdr devices, Copyright (c) 2012, Steve Markgraf
Usage:
GSM Base Station Scan:
kal <-s band indicator> [options]
Clock Offset Calculation:
kal <-f frequency | -c channel> [options]
Where options are:
-s band to scan (GSM850, GSM-R, GSM900, EGSM, DCS, PCS)
-f frequency of nearby GSM base station
-c channel of nearby GSM base station
-b band indicator (GSM850, GSM-R, GSM900, EGSM, DCS, PCS)
-g gain in dB
-d rtl-sdr device index
-e initial frequency error in ppm
-v verbose
-D enable debug messages
-h help
$
$ date ; kal -s GSM850 ; date
Mon Mar 2 16:05:26 CET 2015
Found 1 device(s):
0: Generic RTL2832U OEM
Using device 0: Generic RTL2832U OEM
Found Rafael Micro R820T tuner
Exact sample rate is: 270833.002142 Hz
kal: Scanning for GSM-850 base stations.
GSM-850:
Mon Mar 2 16:08:44 CET 2015
$ date ; kal -s GSM-R ; date
Mon Mar 2 16:16:37 CET 2015
Found 1 device(s):
0: Generic RTL2832U OEM
Using device 0: Generic RTL2832U OEM
Found Rafael Micro R820T tuner
Exact sample rate is: 270833.002142 Hz
kal: Scanning for GSM-R-900 base stations.
GSM-R-900:
Mon Mar 2 16:17:09 CET 2015
$ date ; kal -s GSM900 ; date
Mon Mar 2 16:19:42 CET2015
Found 1 device(s):
0: Generic RTL2832U OEM
Using device 0: Generic RTL2832U OEM
Found Rafael Micro R820T tuner
Exact sample rate is: 270833.002142 Hz
kal: Scanning for GSM-900 base stations.
GSM-900:
chan: 9 (936.8MHz + 22.294kHz) power: 52744.70
chan: 11 (937.2MHz + 22.198kHz) power: 148897.40
chan: 48 (944.6MHz + 21.298kHz) power: 64839.80
chan: 50 (945.0MHz + 21.360kHz) power: 76421.47
chan: 66 (948.2MHz + 21.020kHz) power: 100342.53
chan: 68 (948.6MHz + 21.109kHz) power: 148650.05
chan: 74 (949.8MHz + 21.369kHz) power: 57399.97
Mon Mar 2 16:23:30 CET 2015
$ date ; kal -s EGSM ; date
Mon Mar 2 16:24:36 CET 2015
Found 1 device(s):
0: Generic RTL2832U OEM
Using device 0: Generic RTL2832U OEM
Found Rafael Micro R820T tuner
Exact sample rate is: 270833.002142 Hz
kal: Scanning for E-GSM-900 base stations.
E-GSM-900:
chan: 9 (936.8MHz + 22.059kHz) power: 55782.16
chan: 11 (937.2MHz + 22.136kHz) power: 141612.90
chan: 48 (944.6MHz + 21.194kHz) power: 60347.62
chan: 49 (944.8MHz + 20.859kHz) power: 52131.68
chan: 50 (945.0MHz + 21.268kHz) power: 79892.10
chan: 66 (948.2MHz + 21.026kHz) power: 95478.05
chan: 68 (948.6MHz + 21.074kHz) power: 142679.30
chan: 74 (949.8MHz + 21.314kHz) power: 73531.29
chan: 985 (927.2MHz + 23.002kHz) power: 79791.55
chan: 988 (927.8MHz + 22.555kHz) power: 64283.53
chan: 992 (928.6MHz + 21.866kHz) power: 57287.97
chan: 1000 (930.2MHz + 22.159kHz) power: 192336.94
Mon Mar 2 16:29:38 CET 2015
$
$ date ; kal -c 1000 ; date
Mon Mar 2 16:37:01 CET 2015
Found 1 device(s):
0: Generic RTL2832U OEM
Using device 0: Generic RTL2832U OEM
Found Rafael Micro R820T tuner
Exact sample rate is: 270833.002142 Hz
kal: Calculating clock frequency offset.
Using E-GSM-900 channel 1000 (930.2MHz)
average [min, max] (range, stddev)
+ 22.137kHz [22112, 22160] (49, 13.404841)
overruns: 0
not found: 0
average absolute error: -23.798 ppm
Mon Mar 2 16:37:23 CET 2015
$
Since there where no overruns and the signal was found every time I know that this is a good result, that I can confidently use this value with this particular RTL-SDR hardware in GRC or SDRSharp. I would round this number and use a PPM offset of -24 (0.0024%), since none of the software supports PPB corrections (yet). Even if there were 1 to 2 overruns or 1 to 5 not-found the PPM value would still probably be valid, it just meant that there was interference during the test or the signal strength of the channel used was not high enough, maybe try the test again using a different channel, that has more power (or less interference).
The Kalibrate software actually uses the GSM signals own inbuilt frequency correction bursts, that are used by mobile phone handsets (which also have cheap quartz crystals inside them) to calculate their ppm frequency correction. The original open source software was written for the Ettus Research USRP hardware and was ported work with the RTL-SDR.
When you buy a 28.8MHz quartz crystal in a can, they are generally not 28,800,000Hz. There are two problems that need to be compensated for by the PPM offset in SDR receiver software. The first is manufacturing tolerance, generally about 100 PPM (0.01%) is the range to keep costs down, so the frequency of a new quartz crystal could be anywhere between 28,797,120Hz and 28,802,880Hz at 20 degrees Celsius. The second problem is that the frequency will change slightly as the device warms up. But the good news is once the device has warmed up it will not drift much, unless the ambient room temperature varies.
So basically you scan the local area for active GSM base station channels with a large amount of power (good signal strength, which will probably be from the base station that is physically closest to your current location). Then kal uses this channels GSM frequency correction bursts to workout the PPM (part per million) offset of the RTL-SDR that you are currently calibrating.
Download, compile and install the kal software
$ sudo apt-get install libtool autoconf automake libfftw3-dev$ git clone https://github.com/steve-m/kalibrate-rtl
$ cd kalibrate-rtl
$ ./bootstrap && CXXFLAGS='-W -Wall -O3'
$ ./configure
$ make
$ sudo make install
Use the kal software to scan GSM base stations for channels with high power
My RTL-SDR only goes from 24MHz to 1750MHz, so I can not use the DCS or PCS GSM system, even if they were active in my area. I know that EGSM is active where I am, but I'll test all the frequencies to time and show how long a scans takes when no GSM is present, as well as how long it takes when GSM base stations are present (about 2-10 minutes). This will also provide the warm up time of about 10 minutes for my RTL-SDR to reach its steady state operating temperature. Some of the GSM down link frequencies are used to send the frequency correction bursts, to calibrate mobile phone handsets, so that the phones know the correction offset to use when TX'ing on the up link frequencies. This allows mobile phone makers to use cheaper hardware in their handsets, and if you are making a few billion of them small savings quickly add up to more profit..- GSM850 UPLINK: 824.2- 849.2MHz DOWNLINK: 869.2- 894.2MHz
- GSM-R UPLINK: 876.0- 915.0MHz DOWNLINK: 921.0- 960.0MHz
- GSM900 UPLINK: 890.0- 915.0MHz DOWNLINK: 935.0- 960.0MHz
- EGSM UPLINK: 880.0- 915.0MHz DOWNLINK: 925.0- 960.0MHz
- DCS UPLINK: 1710.2-1784.8MHz DOWNLINK: 1805.2-1879.8MHz
- PCS UPLINK: 1850.2-1909.8MHz DOWNLINK: 1930.2-1989.8MHz
(confirm that the RTL-SDR is actually plugged into the Linux machine)
$ lsusb | grep -i RTL
Bus 004 Device 015: ID 0bda:2838 Realtek Semiconductor Corp. RTL2838 DVB-T
(print the help for the kal software to see what arguments should be used)
$ kal
error: must enter channel or frequency
kalibrate v0.4.1-rtl, Copyright (c) 2010, Joshua Lackey
modified for use with rtl-sdr devices, Copyright (c) 2012, Steve Markgraf
Usage:
GSM Base Station Scan:
kal <-s band indicator> [options]
Clock Offset Calculation:
kal <-f frequency | -c channel> [options]
Where options are:
-s band to scan (GSM850, GSM-R, GSM900, EGSM, DCS, PCS)
-f frequency of nearby GSM base station
-c channel of nearby GSM base station
-b band indicator (GSM850, GSM-R, GSM900, EGSM, DCS, PCS)
-g gain in dB
-d rtl-sdr device index
-e initial frequency error in ppm
-v verbose
-D enable debug messages
-h help
$
$ date ; kal -s GSM850 ; date
Mon Mar 2 16:05:26 CET 2015
Found 1 device(s):
0: Generic RTL2832U OEM
Using device 0: Generic RTL2832U OEM
Found Rafael Micro R820T tuner
Exact sample rate is: 270833.002142 Hz
kal: Scanning for GSM-850 base stations.
GSM-850:
Mon Mar 2 16:08:44 CET 2015
$ date ; kal -s GSM-R ; date
Mon Mar 2 16:16:37 CET 2015
Found 1 device(s):
0: Generic RTL2832U OEM
Using device 0: Generic RTL2832U OEM
Found Rafael Micro R820T tuner
Exact sample rate is: 270833.002142 Hz
kal: Scanning for GSM-R-900 base stations.
GSM-R-900:
Mon Mar 2 16:17:09 CET 2015
$ date ; kal -s GSM900 ; date
Mon Mar 2 16:19:42 CET2015
Found 1 device(s):
0: Generic RTL2832U OEM
Using device 0: Generic RTL2832U OEM
Found Rafael Micro R820T tuner
Exact sample rate is: 270833.002142 Hz
kal: Scanning for GSM-900 base stations.
GSM-900:
chan: 9 (936.8MHz + 22.294kHz) power: 52744.70
chan: 11 (937.2MHz + 22.198kHz) power: 148897.40
chan: 48 (944.6MHz + 21.298kHz) power: 64839.80
chan: 50 (945.0MHz + 21.360kHz) power: 76421.47
chan: 66 (948.2MHz + 21.020kHz) power: 100342.53
chan: 68 (948.6MHz + 21.109kHz) power: 148650.05
chan: 74 (949.8MHz + 21.369kHz) power: 57399.97
Mon Mar 2 16:23:30 CET 2015
$ date ; kal -s EGSM ; date
Mon Mar 2 16:24:36 CET 2015
Found 1 device(s):
0: Generic RTL2832U OEM
Using device 0: Generic RTL2832U OEM
Found Rafael Micro R820T tuner
Exact sample rate is: 270833.002142 Hz
kal: Scanning for E-GSM-900 base stations.
E-GSM-900:
chan: 9 (936.8MHz + 22.059kHz) power: 55782.16
chan: 11 (937.2MHz + 22.136kHz) power: 141612.90
chan: 48 (944.6MHz + 21.194kHz) power: 60347.62
chan: 49 (944.8MHz + 20.859kHz) power: 52131.68
chan: 50 (945.0MHz + 21.268kHz) power: 79892.10
chan: 66 (948.2MHz + 21.026kHz) power: 95478.05
chan: 68 (948.6MHz + 21.074kHz) power: 142679.30
chan: 74 (949.8MHz + 21.314kHz) power: 73531.29
chan: 985 (927.2MHz + 23.002kHz) power: 79791.55
chan: 988 (927.8MHz + 22.555kHz) power: 64283.53
chan: 992 (928.6MHz + 21.866kHz) power: 57287.97
chan: 1000 (930.2MHz + 22.159kHz) power: 192336.94
Mon Mar 2 16:29:38 CET 2015
$
Use the kal software to calculate the PPM clock offset of a RTL-SDR.
So now that I have at least one channel number, with a high broadcast power in my region, I can calculate the PPM offset with the kal software (best if the RTL-SDR has had a chance to warm up, the cheap 28.8MHz quartz crystal probably drifts most with temperature).$ date ; kal -c 1000 ; date
Mon Mar 2 16:37:01 CET 2015
Found 1 device(s):
0: Generic RTL2832U OEM
Using device 0: Generic RTL2832U OEM
Found Rafael Micro R820T tuner
Exact sample rate is: 270833.002142 Hz
kal: Calculating clock frequency offset.
Using E-GSM-900 channel 1000 (930.2MHz)
average [min, max] (range, stddev)
+ 22.137kHz [22112, 22160] (49, 13.404841)
overruns: 0
not found: 0
average absolute error: -23.798 ppm
Mon Mar 2 16:37:23 CET 2015
$
Since there where no overruns and the signal was found every time I know that this is a good result, that I can confidently use this value with this particular RTL-SDR hardware in GRC or SDRSharp. I would round this number and use a PPM offset of -24 (0.0024%), since none of the software supports PPB corrections (yet). Even if there were 1 to 2 overruns or 1 to 5 not-found the PPM value would still probably be valid, it just meant that there was interference during the test or the signal strength of the channel used was not high enough, maybe try the test again using a different channel, that has more power (or less interference).
The Kalibrate software actually uses the GSM signals own inbuilt frequency correction bursts, that are used by mobile phone handsets (which also have cheap quartz crystals inside them) to calculate their ppm frequency correction. The original open source software was written for the Ettus Research USRP hardware and was ported work with the RTL-SDR.
Tuesday, 25 November 2014
Banana Pi USB 2.0 'Gigabit' Ethernet Adapter throughput test
Banana Pi USB 2.0 'Gigabit' Ethernet Adapter throughput test
I bought a cheap USB 2.0 to Gigabit Ethernet Adapter (Maximum Data Transfer Rate
480 Mbps according to the website), mostly because I wanted to see how it performed on a Raspberry Pi, so now I'm running the same tests on my new board the Banana Pi. It can never reach 480Mbps, but it will be interesting to see how much throughput it can actually achieve. As usual, I have censored any identifying information about my local network or personal hardware.
Hardware Configuration:
Banana PiRAM 1024MB
no overclocking
Inbuilt 10/100/1000 NIC is not connected.
Top USB port has USB 2.0 to Gigabit Ethernet Adapter plugged in and plugged.into router
Bottom USB port is empty
+---+ +------+ +------+ +----------------+
|BPi|->|USB2.0|->1Gbps->|router|->200Mbps->|Remote webserver|
+---+ +------+ Ethernet+------+ Internet +----------------+
480Mibps |
1Gbps Ethernet
|
v
+-------------+
|LAN webserver|
+-------------+
Software Configuration:
I had to add in an extra three lines for this new USB eth1 NIC to get it working on the Banana Pi, the lines in bold below.root@bananapi ~ # cat /etc/network/interfaces
allow-hotplug eth1
auto lo
iface lo inet loopback
auto eth0
iface eth0 inet dhcp
auto eth1
iface eth1 inet dhcp
root@bananapi ~ #
root@bananapi ~ # lsusb
Bus 001 Device 001: ID 1d6b:0002 Linux Foundation 2.0 root hub
Bus 005 Device 001: ID 1d6b:0001 Linux Foundation 1.1 root hub
Bus 003 Device 001: ID 1d6b:0001 Linux Foundation 1.1 root hub
Bus 004 Device 002: ID 0424:7500 Standard Microsystems Corp. LAN7500 Ethernet 10/100/1000 Adapter
Bus 004 Device 001: ID 1d6b:0002 Linux Foundation 2.0 root hub
Bus 002 Device 001: ID 1d6b:0002 Linux Foundation 2.0 root hub
root@bananapi ~ # lsusb -t
/: Bus 05.Port 1: Dev 1, Class=root_hub, Driver=sw-ohci/1p, 12M
/: Bus 04.Port 1: Dev 1, Class=root_hub, Driver=sw-ehci/1p, 480M
|__ Port 1: Dev 2, If 0, Class=Vendor Specific Class, Driver=smsc75xx, 480M
/: Bus 03.Port 1: Dev 1, Class=root_hub, Driver=sw-ohci/1p, 12M
/: Bus 02.Port 1: Dev 1, Class=root_hub, Driver=sw-ehci/1p, 480M
/: Bus 01.Port 1: Dev 1, Class=root_hub, Driver=sw_hcd_host0/1p, 480M
root@bananapi ~ # ethtool eth1
Settings for eth1:
Supported ports: [ TP MII ]
Supported link modes: 10baseT/Half 10baseT/Full
100baseT/Half 100baseT/Full
1000baseT/Half 1000baseT/Full
Supported pause frame use: No
Supports auto-negotiation: Yes
Advertised link modes: 10baseT/Half 10baseT/Full
100baseT/Half 100baseT/Full
1000baseT/Full
Advertised pause frame use: Symmetric Receive-only
Advertised auto-negotiation: Yes
Link partner advertised link modes: 10baseT/Half 10baseT/Full
100baseT/Half 100baseT/Full
1000baseT/Half 1000baseT/Full
Link partner advertised pause frame use: Symmetric Receive-only
Link partner advertised auto-negotiation: Yes
Speed: 1000Mb/s
Duplex: Full
Port: MII
PHYAD: 1
Transceiver: internal
Auto-negotiation: on
Current message level: 0x00000007 (7)
drv probe link
Link detected: yes
root@bananapi ~ #
root@bananapi ~ # netstat -i
Kernel Interface table
Iface MTU Met RX-OK RX-ERR RX-DRP RX-OVR TX-OK TX-ERR TX-DRP TX-OVR Flg
eth0 1500 0 0 0 0 0 4 0 0 0 BMU
eth1 1500 0 457 0 0 0 485 0 0 0 BMRU
lo 16436 0 1 0 0 0 1 0 0 0 LRU
root@bananapi ~ #
root@bananapi ~ # ifconfig eth1
eth1 Link encap:Ethernet HWaddr 80:3f:5d:**:**:**
inet addr:*.*.*.* Bcast:*.*.*.* Mask:*.*.*.*
inet6 addr: fe80::823f:5dff:fe08:69ad/64 Scope:Link
UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1
RX packets:470 errors:0 dropped:0 overruns:0 frame:0
TX packets:502 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:1000
RX bytes:37207 (36.3 KiB) TX bytes:72520 (70.8 KiB)
root@bananapi ~ #
Internet Test (via a 200Mbit/sec broadband connection)
Download a single 100MiB file to the Pi through the USB 2.0 Gigabit Ethernet Adapter and send the downloaded data to /dev/null.
root@bananapi ~ # time wget http://qrng.physik.hu-berlin.de/files/speedtest-100MB.bin -O /dev/null
--2010-01-01 00:17:55-- http://qrng.physik.hu-berlin.de/files/speedtest-100MB.bin
Resolving qrng.physik.hu-berlin.de (qrng.physik.hu-berlin.de)... 141.20.41.134
Connecting to qrng.physik.hu-berlin.de (qrng.physik.hu-berlin.de)|141.20.41.134|:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 104857600 (100M) [application/octet-stream]
Saving to: ‘/dev/null’
/dev/null 100%[=====================================================================================================================>] 100.00M 11.4MB/s in 16s
2010-01-01 00:18:11 (6.21 MB/s) - ‘/dev/null’ saved [104857600/104857600]
wget http://qrng.physik.hu-berlin.de/files/speedtest-100MB.bin -O /dev/null 0.25s user 2.22s system 14% cpu 16.559 total
root@bananapi ~ #
LAN Test
Download a single 100MiB file to the Pi through USB 2.0 Gigabit Ethernet Adapter and send the downloaded data to /dev/null. The data source is on the Local network from a machine with a 1 Gbit/sec NIC patched directly into the router with the data being read from a RAM disk to maximise read speed.root@webserver:~# apt-get install nginx nginx-common nginx-full
root@webserver:~# /usr/sbin/nginx &
root@webserver:~# mkdir /usr/share/nginx/www/ramdisk
root@webserver:~# chmod 777 /usr/share/nginx/www/ramdisk
root@webserver:~# free -m
root@webserver:~# mount -t tmpfs -o size=256M tmpfs /usr/share/nginx/www/ramdisk
root@webserver:~# cp speedtest-100MB.bin /usr/share/nginx/www/ramdisk
root@bananapi ~ # time wget http://*.*.*.*/ramdisk/speedtest-100MB.bin -O /dev/null
--2014-11-20 21:10:47-- http://*.*.*.*/ramdisk/speedtest-100MB.bin
Connecting to *.*.*.*:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 104857600 (100M) [application/octet-stream]
Saving to: ‘/dev/null’
/dev/null 100%[=====================================================================================================================>] 100.00M 39.5MB/s in 2.5s
2014-11-20 21:10:50 (39.5 MB/s) - ‘/dev/null’ saved [104857600/104857600]
wget http://*.*.*.*/ramdisk/speedtest-100MB.bin -O /dev/null 0.24s user 1.60s system 71% cpu 2.572 total
root@bananapi ~ #
Results
Peak Internet transfer rate 10.87MiB/sec (91.2Mbps) - about 13 hops away
Average Internet throughput 6.04MiB/sec - about 13 hops away.
Because this server is so many hops away it's throughput varies wildly with time of day and is effected by so many uncontrolled fluctuations on every hop in between. This result is more an example of a typical Internet based throughput than a maximum throughput.
Peak LAN download transfer rate 38.88 MiB/sec (326.15Mbps) - for 1000Mbps (even limited to USB 2.0 480Mbps) this is a very good throughput.
Average LAN download throughput 38.88 MiB/sec - for 1000Mbps (even limited to USB 2.0 480Mbps) this is a very good throughput.
I'm very pleased with the USB 2.0 Gigabit Ethernet Adapter result for the Banana Pi. I was expecting a result somewhere between 20 MiB/sec and 40 MiB/sec.
I enabled jumbo frames ("sudo ifconfig eth1 mtu 9000") on the webserver and on the Banana Pi to see if I could get a better result, but I couldn't get it to work. I'm not quite sure why, yet but I'm still looking into it.
Additional testing
I decided to try the test in reverse. Install a webserver on the Banana P, and make a 128MiB ramdisk to download a file from the Banana Pi through USB 2.0, through the USB 2.0 Gigabit Ethernet Adapter down to the midrange PC with a Gigabit NIC, and then throw the data away.root@bananapi ~ # apt-get install nginx
root@bananapi ~ # mkdir /var/www/html/ramdisk
root@bananapi ~ # chmod 777 /var/www/html/ramdiskroot@bananapi ~ # free -m
root@bananapi ~ # mount -t tmpfs -o size=128M tmpfs /var/www/html/ramdisk
root@bananapi ~ # cp speedtest-100MB.bin /var/www/html/ramdisk
midrangepc $ time wget http://*.*.*.*/ramdisk/speedtest-100MB.bin -O /dev/null
--2014-11-20 22:39:33-- http://*.*.*.*/ramdisk/speedtest-100MB.bin
Connecting to *.*.*.*:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 104857600 (100M) [application/octet-stream]
Saving to: `/dev/null'
100%[=================================================================================================================================================================>] 104,857,600 19.8MB/s in 5.1s
2014-11-20 22:39:38 (19.7 MB/s) - `/dev/null' saved [104857600/104857600]
real 0m5.075s
user 0m0.268s
sys 0m1.592smidrangepc $
Average LAN upload throughput 19.6 MiB/sec (156.8Mbps)
I'm not quite sure why upload is always less than download throughput rates, it is something that I'll have to think about for a while.
For two final tests I decided to unplug the USB 2.0 Gigabit ethernet adapter and use the Banana Pi's actual builtin Gigabit NIC and just get the average throughput in each direction (using a ramdisk on each machine to upload the file from as fast as possible).
Midrange server as the webserver: 101 MB/s (Banana Pi as the client, running wget) [808Mbps 96.3MiB/sec]
Banana Pi as the webserver: 71.5 MB/s (Midrange PC server as the client, running wget) [572Mbps 68.2MiB/sec]
Labels:
1000Mbps,
Banana Pi,
Ethernet,
Gigabit,
throughput,
USB 2.0,
USB21000S2
Thursday, 20 November 2014
Comparison of Raspberry Pi vs Beaglebone Black vs Banana Pi
Update (2015-2): I'll update the below when I get more info on the new BCM2836 chip used in the Raspberry Pi Generation 2 model B. It CPU performance should jump up by about six, the RAM has doubled, but the GPU is the same as is everything else. RAM performance will be the same, no clocking improvement, just size. The quad core CPU (ARMv7) does now have NEON support so SIMD, should improve data processing ability. But the real bottleneck to the RPi is getting data into and out of the CPU from the real world, it is like a giant head sitting on a straw.
Raspberry Pi (Model A,B,B+,A+ all have the same ARM1176JZF-S)
------------
Processor : ARMv6-compatible processor rev 7 (v6l)
BogoMIPS : 697.95
Features : swp half thumb fastmult vfp edsp java tls
CPU implementer : 0x41
CPU architecture: 7
CPU variant : 0x0
CPU part : 0xb76
CPU revision : 7
Hardware : BCM2708
Revision : 000f
BeagleBone Black
----------------
processor : 0
model name : ARMv7 Processor rev 2 (v7l)
BogoMIPS : 990.68
Features : swp half thumb fastmult vfp edsp thumbee neon vfpv3 tls
CPU implementer : 0x41
CPU architecture: 7
CPU variant : 0x3
CPU part : 0xc08
CPU revision : 2
Hardware : Generic AM33XX (Flattened Device Tree)
Revision : 0000
Serial : 0000000000000000
Banana Pi (The Banana Pro has the same Allwinner A20)
---------
Processor : ARMv7 Processor rev 4 (v7l)
processor : 0
BogoMIPS : 2004.17
processor : 1
BogoMIPS : 2011.05
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
Raspberry Pi: Broadcom VideoCore IV (Graphics and compute) 24 GFLOPS
Beaglebone Black: PowerVR SGX530 (200 MHz) (Graphics, video encoding, decoding only) 1.6 GFLOPS
Banana Pi: ARM Mali400MP2 dual GPU core (Grahpics, video encoding, decoding only) 6.3 GFLOPS
Raspberry Pi ( A, A+, B rev 1): 256MiB DDR
Raspberry Pi (B rev 2, B+): 512MiB DDR
BeagleBone Black: 512MiB DDR3L
Banana Pi: 1024MiB DDR3
[DDR3 is 1.35 volt and DDR3L 1.25 volt (lower power = less heat = longer life)]
Raspberry Pi A: 1 x Standard A host port (direct).
Raspberry Pi B: 2 x Standard A host port (via hub with Ethernet).
Raspberry Pi A+: 1 x Standard A host port (direct).
Raspberry Pi B+: 4 x Standard A host port (via hub with Ethernet).
Beaglebone Black: 1 x Standard A host port (direct) 1x mini B device port (direct)
Banana Pi: 2 x Standard A host port (direct) 1x mini AB OTG port (direct)
Beaglebone Black: 10/100Mbps
Banana Pi: 10/100/1000Mbps
Raspberry Pi A+, B+: microSD card
Beaglebone Black: microSD card
Banana Pi: SD card
Beaglebone Black ((Rev B): 2 GB Ångström pre-installed
Beaglebone Black ((Rev C): 4 GB Debian pre-installed
Banana Pi: None
Beaglebone Black: None
Banana Pi: SATA 2.0
Me personally I hate devices with inbuilt flash, 3000-5000 writes per block and the device ready for the bin.
8-17GPIO pins
Raspberry Pi B rev 2: UART, SPI, 2x I²C,MIPI CSI-2, (not yet enabled DSI, CEC)
12-21GPIO pins
Raspberry Pi A+, B+: UART, SPI, 2x I²C,MIPI CSI-2, (not yet enabled DSI, CEC)
EEPROM ID feature for auto-configuration with add-on "HAT" boards
21-30GPIO pins
Beaglebone Black: 4xUART(1xTX only), 8x PWM, LCD, GPMC, MMC1, 2x SPI, 2x I²C, 7xADC(1.8v), 2x CAN bus, 4 Timers, 25xPRU
65 GPIO pins
Banana Pi: UART, SPI, I²C, CAN, ADC, PWM
CSI (or 21 additional GPIO pins), LCD display LVDS (or 36 additional GPIO pins)
7-17 GPIO pins (or 7-74 GPIO pins if CSI and LCD are re-purposed)
IR receiver
On board microphone
Overall I'd probably pick the Beaglebone Black as having the best GPIO/Low-level peripherals support (and very easy access to the pins).
Raspberry Pi B: 700-1000mA
Raspberry Pi A+: 500mA-2000mA
Raspberry Pi B+: 600-2000mA
Beaglebone Black: 300–500 mA @5 V
Banana Pi: 200mA*-2000mA (*requires 700-800mA during boot)
Raspberry Pi B: 45 g (1.6 oz)
Raspberry Pi A+: 23 g (0.81 oz)
Raspberry Pi B+: 45 g (1.6 oz)
Beaglebone Black: 39.68 g (1.400 oz)
Banana Pi: 48 g (1.7 oz)
My conclusion would be if it is light weight I needed, then I'd go with a Pro Trinket (2.6g/0.09oz), unless I needed a 1920x1080HD camera as well then it would be the Raspberry Pi A+ (23g/0.81oz) with a camera module (3g/0.11oz). If it was easily accessible GPIO that I needed, with loads of inputs and outputs I'd pick the Beaglebone Black (It is a great piece of open hardware) 7ADC is nice. And finally if it was shoving large amounts of data about fast, or processing that data, I'd pick the Banana Pi (With SATA 2.0, Gigabit Ethernet and 1GiB of RAM it is an impressive piece of kit) or it's newer incarnation of hardware the Banana Pro (with builtin WiFi).
There are larger communities built up around the Raspberry Pi and Beaglebone Black, than there is around the Banana Pi/Pro hardware. So if you have a problem, sometimes you may have to solve it yourself rather than looking for a quick fix from someone else in the community who had the same problem already.
CPU
From each of the /proc/cpuinfo files shown below, in theory if an application is single threaded the Banana Pi should win, and if it is multithreaded the Banana Pi should still win. I have not run any direct benchmark, just looking at the BogoMIPS and the Features supported by each ARM chip.Raspberry Pi (Model A,B,B+,A+ all have the same ARM1176JZF-S)
------------
Processor : ARMv6-compatible processor rev 7 (v6l)
BogoMIPS : 697.95
Features : swp half thumb fastmult vfp edsp java tls
CPU implementer : 0x41
CPU architecture: 7
CPU variant : 0x0
CPU part : 0xb76
CPU revision : 7
Hardware : BCM2708
Revision : 000f
BeagleBone Black
----------------
processor : 0
model name : ARMv7 Processor rev 2 (v7l)
BogoMIPS : 990.68
Features : swp half thumb fastmult vfp edsp thumbee neon vfpv3 tls
CPU implementer : 0x41
CPU architecture: 7
CPU variant : 0x3
CPU part : 0xc08
CPU revision : 2
Hardware : Generic AM33XX (Flattened Device Tree)
Revision : 0000
Serial : 0000000000000000
Banana Pi (The Banana Pro has the same Allwinner A20)
---------
Processor : ARMv7 Processor rev 4 (v7l)
processor : 0
BogoMIPS : 2004.17
processor : 1
BogoMIPS : 2011.05
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
Video/GPU
Raspberry Pi: Broadcom VideoCore IV (Graphics and compute) 24 GFLOPS
Beaglebone Black: PowerVR SGX530 (200 MHz) (Graphics, video encoding, decoding only) 1.6 GFLOPS
Banana Pi: ARM Mali400MP2 dual GPU core (Grahpics, video encoding, decoding only) 6.3 GFLOPS
RAM
Raspberry Pi ( A, A+, B rev 1): 256MiB DDR
Raspberry Pi (B rev 2, B+): 512MiB DDR
BeagleBone Black: 512MiB DDR3L
Banana Pi: 1024MiB DDR3
[DDR3 is 1.35 volt and DDR3L 1.25 volt (lower power = less heat = longer life)]
USB
Raspberry Pi A: 1 x Standard A host port (direct).
Raspberry Pi B: 2 x Standard A host port (via hub with Ethernet).
Raspberry Pi A+: 1 x Standard A host port (direct).
Raspberry Pi B+: 4 x Standard A host port (via hub with Ethernet).
Beaglebone Black: 1 x Standard A host port (direct) 1x mini B device port (direct)
Banana Pi: 2 x Standard A host port (direct) 1x mini AB OTG port (direct)
Ethernet
Raspberry Pi (B, B+): 10/100Mbps (shared USB)Beaglebone Black: 10/100Mbps
Banana Pi: 10/100/1000Mbps
Storage
SD/microSD
Raspberry Pi A,B: SD cardRaspberry Pi A+, B+: microSD card
Beaglebone Black: microSD card
Banana Pi: SD card
8-bit eMMC
Raspberry Pi: NoneBeaglebone Black ((Rev B): 2 GB Ångström pre-installed
Beaglebone Black ((Rev C): 4 GB Debian pre-installed
Banana Pi: None
SATA
Raspberry Pi: NoneBeaglebone Black: None
Banana Pi: SATA 2.0
Me personally I hate devices with inbuilt flash, 3000-5000 writes per block and the device ready for the bin.
GPIO/Low-level peripherals
Raspberry Pi A, B rev 1: UART, SPI, 2x I²C,MIPI CSI-2, (not yet enabled DSI, CEC)8-17GPIO pins
Raspberry Pi B rev 2: UART, SPI, 2x I²C,MIPI CSI-2, (not yet enabled DSI, CEC)
12-21GPIO pins
Raspberry Pi A+, B+: UART, SPI, 2x I²C,MIPI CSI-2, (not yet enabled DSI, CEC)
EEPROM ID feature for auto-configuration with add-on "HAT" boards
21-30GPIO pins
Beaglebone Black: 4xUART(1xTX only), 8x PWM, LCD, GPMC, MMC1, 2x SPI, 2x I²C, 7xADC(1.8v), 2x CAN bus, 4 Timers, 25xPRU
65 GPIO pins
Banana Pi: UART, SPI, I²C, CAN, ADC, PWM
CSI (or 21 additional GPIO pins), LCD display LVDS (or 36 additional GPIO pins)
7-17 GPIO pins (or 7-74 GPIO pins if CSI and LCD are re-purposed)
IR receiver
On board microphone
Overall I'd probably pick the Beaglebone Black as having the best GPIO/Low-level peripherals support (and very easy access to the pins).
Power
Raspberry Pi A: 500mA-1000mARaspberry Pi B: 700-1000mA
Raspberry Pi A+: 500mA-2000mA
Raspberry Pi B+: 600-2000mA
Beaglebone Black: 300–500 mA @5 V
Banana Pi: 200mA*-2000mA (*requires 700-800mA during boot)
Weight
Raspberry Pi A: 45 g (1.6 oz)Raspberry Pi B: 45 g (1.6 oz)
Raspberry Pi A+: 23 g (0.81 oz)
Raspberry Pi B+: 45 g (1.6 oz)
Beaglebone Black: 39.68 g (1.400 oz)
Banana Pi: 48 g (1.7 oz)
My conclusion would be if it is light weight I needed, then I'd go with a Pro Trinket (2.6g/0.09oz), unless I needed a 1920x1080HD camera as well then it would be the Raspberry Pi A+ (23g/0.81oz) with a camera module (3g/0.11oz). If it was easily accessible GPIO that I needed, with loads of inputs and outputs I'd pick the Beaglebone Black (It is a great piece of open hardware) 7ADC is nice. And finally if it was shoving large amounts of data about fast, or processing that data, I'd pick the Banana Pi (With SATA 2.0, Gigabit Ethernet and 1GiB of RAM it is an impressive piece of kit) or it's newer incarnation of hardware the Banana Pro (with builtin WiFi).
There are larger communities built up around the Raspberry Pi and Beaglebone Black, than there is around the Banana Pi/Pro hardware. So if you have a problem, sometimes you may have to solve it yourself rather than looking for a quick fix from someone else in the community who had the same problem already.
Labels:
1000Mbps,
100Mbps,
Allwinner A20,
ARM1176JZF-S,
Banana Pi,
BCM2835,
BCM2836,
Beaglebone Black,
Ethernet,
flash,
Gigabit,
Linux,
Raspberry Pi,
SATA 2.0,
TI AM335X,
USB 2.0
Wednesday, 12 November 2014
Banana Pi SATA 2.0 disk throughput test
Hardware Configuration:
Banana Pi-M1 1024MiB RAM, no overclockingInbuilt 10/100/1000 NIC is connected.
Top USB port is empty.
Bottom USB port is empty.
SATA disk connected via SATA cable to SATA port on BPi.
Harddisk is power from an external supply (R-Driver III).
No keyboard and no mouse.
The HDD that I'm using has an Internal write rate of about 100MiB/sec, so it is not a great disk for SATA performance testing, but it is the same make and model that I used for testing the Raspberry Pi USB disk throughput test.
Prerequisites:
apt-get install dosfstoolsapt-get install expect expect-dev tcl8.6 tk8.6
ext2, ext3, ext4, vfat file system write tests
I split a 2 TB disk into four equal sized partitions.
root@bananapi ~ # fdisk -l /dev/sda
Disk /dev/sda: 1.8 TiB, 2000398934016 bytes, 3907029168 sectors
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 4096 bytes
I/O size (minimum/optimal): 4096 bytes / 4096 bytes
Disklabel type: dos
Disk identifier: 0x5d3d4ef6
Device Boot Start End Sectors Size Id Type
/dev/sda1 2048 976758783 976756736 465.8G 83 Linux
/dev/sda2 976758784 1953515519 976756736 465.8G 83 Linux
/dev/sda3 1953515520 2930272255 976756736 465.8G 83 Linux
/dev/sda4 2930272256 3907028991 976756736 465.8G c W95 FAT32 (LBA)
root@bananapi ~ # mkfs.ext2 -L ext2 /dev/sda1
mke2fs 1.42.12 (29-Aug-2014)
Creating filesystem with 122094592 4k blocks and 30531584 inodes
Filesystem UUID: 35d529af-5e67-4f17-8c78-5af67a4f85f8
Superblock backups stored on blocks:
32768, 98304, 163840, 229376, 294912, 819200, 884736, 1605632, 2654208, 4096000, 7962624, 11239424, 20480000, 23887872, 71663616, 78675968, 102400000
Allocating group tables: done
Writing inode tables: done
Writing superblocks and filesystem accounting information: done
mkfs.ext2 -L ext2 /dev/sda1 0.65s user 31.41s system 12% cpu 4:18.17 total
root@bananapi ~ # mkfs.ext3 -L ext3 /dev/sda2
mke2fs 1.42.12 (29-Aug-2014)
Creating filesystem with 122094592 4k blocks and 30531584 inodes
Filesystem UUID: d0fe6f90-cc02-4d98-ba62-a704692edaf3
Superblock backups stored on blocks:
32768, 98304, 163840, 229376, 294912, 819200, 884736, 1605632, 2654208, 4096000, 7962624, 11239424, 20480000, 23887872, 71663616, 78675968, 102400000
Allocating group tables: done
Writing inode tables: done
Creating journal (32768 blocks): done
Writing superblocks and filesystem accounting information: done
mkfs.ext3 -L ext2 /dev/sda2 0.81s user 33.67s system 13% cpu 4:20.11 total
root@bananapi ~ # mkfs.ext4 -L ext4 /dev/sda3
mke2fs 1.42.12 (29-Aug-2014)
Creating filesystem with 122094592 4k blocks and 30531584 inodes
Filesystem UUID: 02029d9e-70e3-4012-a217-fff6754fe0b8
Superblock backups stored on blocks:
32768, 98304, 163840, 229376, 294912, 819200, 884736, 1605632, 2654208, 4096000, 7962624, 11239424, 20480000, 23887872, 71663616, 78675968, 102400000
Allocating group tables: done
Writing inode tables: done
Creating journal (32768 blocks): done
Writing superblocks and filesystem accounting information: done
root@bananapi ~ # mkfs.vfat -n VFAT /dev/sda4
mkfs.fat 3.0.26 (2014-03-07)
root@bananapi ~ # cd /media
root@bananapi /media # mkdir ext2 ext3 ext4 vfat
root@bananapi /media # cd
root@bananapi ~ # cat write-throughputtest.exp
#!/usr/bin/expect -f
set force_conservative 0 ;# set to 1 to force conservative mode even if
;# script wasn't run conservatively originally
if {$force_conservative} {
set send_slow {1 .1}
proc send {ignore arg} {
sleep .1
exp_send -s -- $arg
}
}
set timeout -1
spawn $env(SHELL)
match_max 100000
expect "root@bananapi"
send -- "script /root/output.txt\r"
expect "Script started, file is /root/output.txt\r"
send -- "mount /dev/sda1 /media/ext2\r"
expect "root@bananapi"
send -- "mount /dev/sda2 /media/ext3\r"
expect "root@bananapi"
send -- "mount /dev/sda3 /media/ext4\r"
expect "root@bananapi"
send -- "mount /dev/sda4 /media/vfat\r"
expect "root@bananapi"
send -- "/root/write-throughputtest.sh /media/ext2\r"
expect "root@bananapi"
send -- "/root/write-throughputtest.sh /media/ext3\r"
expect "root@bananapi"
send -- "/root/write-throughputtest.sh /media/ext4\r"
expect "root@bananapi"
send -- "/root/write-throughputtest.sh /media/vfat\r"
expect "root@bananapi"
send -- "exit\r"
expect "Script done, file is /root/output.txt\r"
send -- "exit\r"
expect eof
root@bananapi ~ #
root@bananapi ~ # cat write-throughputtest.sh
#!/bin/sh
DEST=$1
# 18 x 1GiB = 18GiB
# assuming maximum write SATA2 286MiB/sec this should take more than a minute
# assuming poor write speed of 5 MiB/sec this should take more than a hour
# either way one pass should happen within 12 hours
echo touch $DEST/testbegin-write-starttime
touch $DEST/testbegin-write-starttime
date
echo sync
sync
date
for BS in 8 16 32 64 128 256 \
512 1024 2048 4096 8192 16384 \
32768 65536 131072 262144 524288 1048576
do
COUNT=`expr 1073741824 / $BS`
echo touch $DEST/${BS}b-write-starttime
touch $DEST/${BS}b-write-starttime
date
echo dd if=/dev/zero of=$DEST/${BS}b-data bs=${BS} count=${COUNT}
dd if=/dev/zero of=$DEST/${BS}b-data bs=${BS} count=${COUNT}
date
echo touch $DEST/${BS}b-write-endtime
touch $DEST/${BS}b-write-endtime
date
echo sync
sync
date
# any delay caused by sync will show up on next starttime
done
echo touch $DEST/testover-write-endtime
touch $DEST/testover-write-endtime
date
root@bananapi ~ #
root@bananapi ~ # cat read-throughputtest.sh
#!/bin/sh
SRCE=$1
# 18 x 1GiB = 18GiB
# assuming maximum read SATA2 286MB/sec this whould take just over a minute
# assuming poor read speed of 5 MiB/sec this should take just over a hour
# This script assumes that write-throughputtest.sh was run
# previously to generate the data files.
date
echo sync
sync
date
for BS in 8 16 32 64 128 256 \
512 1024 2048 4096 8192 16384 \
32768 65536 131072 262144 524288 1048576
do
COUNT=`expr 1073741824 / $BS`
date
echo dd of=/dev/null if=$SRCE/${BS}b-data bs=${BS} count=${COUNT}
dd of=/dev/null if=$SRCE/${BS}b-data bs=${BS} count=${COUNT}
date
echo sync
sync
date
done
date
root@bananapi ~ #
I'm converting the results so that I can display them in MiB/sec instead of dd's default of MB/sec. Because I'm copying a Gigabyte of data with each block size all I need to do is divide 1024 by the number of seconds that it took to get the result in MiB/second.
root@bananapi ~ # grep bytes output.txt | grep copied | head -18
1073741824 bytes (1.1 GB) copied, 775.481 s, 1.4 MB/s
1073741824 bytes (1.1 GB) copied, 428.811 s, 2.5 MB/s
1073741824 bytes (1.1 GB) copied, 222.595 s, 4.8 MB/s
1073741824 bytes (1.1 GB) copied, 115.167 s, 9.3 MB/s
1073741824 bytes (1.1 GB) copied, 67.9903 s, 15.8 MB/s
1073741824 bytes (1.1 GB) copied, 43.2929 s, 24.8 MB/s
1073741824 bytes (1.1 GB) copied, 28.109 s, 38.2 MB/s
1073741824 bytes (1.1 GB) copied, 24.2247 s, 44.3 MB/s
1073741824 bytes (1.1 GB) copied, 23.847 s, 45.0 MB/s
1073741824 bytes (1.1 GB) copied, 23.8354 s, 45.0 MB/s
1073741824 bytes (1.1 GB) copied, 23.7234 s, 45.3 MB/s
1073741824 bytes (1.1 GB) copied, 23.8386 s, 45.0 MB/s
1073741824 bytes (1.1 GB) copied, 23.6681 s, 45.4 MB/s
1073741824 bytes (1.1 GB) copied, 24.1159 s, 44.5 MB/s
1073741824 bytes (1.1 GB) copied, 24.3767 s, 44.0 MB/s
1073741824 bytes (1.1 GB) copied, 24.477 s, 43.9 MB/s
1073741824 bytes (1.1 GB) copied, 24.8436 s, 43.2 MB/s
1073741824 bytes (1.1 GB) copied, 24.3786 s, 44.0 MB/s
root@bananapi ~ # grep bytes output.txt | grep copied | awk '{printf "%3.1f\n", 1024/$6}' | head -18
1.3
2.4
4.6
8.9
15.1
23.7
36.4
42.3
42.9
43.0
43.2
43.0
43.3
42.5
42.0
41.8
41.2
42.0
root@bananapi ~ #
Results
SATA disk performance check on Banana Pi SATA port.
Banana Pi SATA disk write results (spinning rust)
-----------------------------------------------------------
block filesystem type
size -------------------------------------------
(bytes)ext2 ext3 ext4 vfat
8 1.3 MiB/s 0.6 MiB/s 0.6 MiB/s 1.1 MiB/s
16 2.4 MiB/s 1.1 MiB/s 1.1 MiB/s 2.1 MiB/s
32 4.6 MiB/s 2.1 MiB/s 2.1 MiB/s 3.9 MiB/s
64 8.9 MiB/s 4.0 MiB/s 4.1 MiB/s 7.3 MiB/s
128 15.1 MiB/s 7.6 MiB/s 7.9 MiB/s 11.9 MiB/s
256 23.7 MiB/s 13.0 MiB/s 12.5 MiB/s 16.7 MiB/s
512 36.4 MiB/s 21.3 MiB/s 23.2 MiB/s 23.0 MiB/s
1k 42.3 MiB/s 30.9 MiB/s 36.0 MiB/s 26.4 MiB/s
2k 42.9 MiB/s 36.2 MiB/s 43.8 MiB/s 28.7 MiB/s
4k 43.0 MiB/s 40.3 MiB/s 44.5 MiB/s 29.7 MiB/s
8k 43.2 MiB/s 41.0 MiB/s 44.2 MiB/s 30.6 MiB/s
16k 43.0 MiB/s 40.6 MiB/s 44.0 MiB/s 31.0 MiB/s
32k 43.3 MiB/s 37.5 MiB/s 44.0 MiB/s 30.2 MiB/s
64k 42.5 MiB/s 39.9 MiB/s 43.4 MiB/s 30.4 MiB/s
128k 42.0 MiB/s 39.9 MiB/s 42.3 MiB/s 30.1 MiB/s
256k 41.8 MiB/s 39.1 MiB/s 43.3 MiB/s 29.8 MiB/s
512k 41.2 MiB/s 38.9 MiB/s 43.0 MiB/s 30.6 MiB/s
1M 42.0 MiB/s 39.6 MiB/s 43.5 MiB/s 30.0 MiB/s
Banana Pi SATA disk read results (spinning rust)
-----------------------------------------------------------
block filesystem type
size -------------------------------------------
(bytes)ext2 ext3 ext4 vfat
8 2.1MiB/s 2.1MiB/s 2.0MiB/s 2.0MiB/s
16 4.0MiB/s 4.1MiB/s 4.1MiB/s 4.0MiB/s
32 8.7MiB/s 8.6MiB/s 8.5MiB/s 8.6MiB/s
64 17.0MiB/s 16.9MiB/s 17.0MiB/s 16.9MiB/s
128 31.3MiB/s 31.2MiB/s 31.5MiB/s 30.5MiB/s
256 54.8MiB/s 54.2MiB/s 55.1MiB/s 52.4MiB/s
512 84.6MiB/s 100.4MiB/s 83.9MiB/s 72.8MiB/s
1k 108.7MiB/s 100.4MiB/s 86.5MiB/s 74.2MiB/s
2k 107.7MiB/s 99.6MiB/s 87.2MiB/s 73.8MiB/s
4k 109.7MiB/s 100.4MiB/s 86.3MiB/s 72.6MiB/s
8k 109.7MiB/s 100.8MiB/s 87.7MiB/s 73.1MiB/s
16k 109.8MiB/s 100.4MiB/s 86.6MiB/s 74.1MiB/s
32k 108.8MiB/s 98.0MiB/s 87.6MiB/s 73.5MiB/s
64k 109.2MiB/s 100.4MiB/s 86.0MiB/s 72.7MiB/s
128k 109.3MiB/s 100.7MiB/s 85.7MiB/s 73.2MiB/s
256k 108.7MiB/s 100.5MiB/s 86.8MiB/s 74.1MiB/s
512k 108.9MiB/s 99.6MiB/s 88.1MiB/s 73.3MiB/s
1M 109.1MiB/s 99.2MiB/s 86.7MiB/s 72.9MiB/s
I decided to run the tests again with an old 128GB SSD (write performance will be OKish because it is used, but read performance should be better than spinning rust). But to avoid killing it (small writes cause SSD's to fail prematurely) I started the write testing at 4K block size.The read test block size does not matter at all, SSD's always fail due to running out of writes (3000-5000 per block) long before any wear due to reads can happen. The reason I ran the tests again with a SSD was because I felt that the limiting factor was the Harddisk's poor read/write performance and not any limit of the Banana Pi's SATA 2.0 port.Banana Pi SATA disk write results (SSD)
-----------------------------------------------------------
block filesystem type
size -------------------------------------------
(bytes)ext2 ext3 ext4 vfat
4k 44.9 MiB/s 40.0 MiB/s 42.5 MiB/s 29.0 MiB/s
8k 45.3 MiB/s 41.0 MiB/s 42.5 MiB/s 29.8 MiB/s
16k 44.2 MiB/s 41.0 MiB/s 42.7 MiB/s 29.5 MiB/s
32k 43.9 MiB/s 41.7 MiB/s 42.4 MiB/s 29.3 MiB/s
64k 43.6 MiB/s 40.4 MiB/s 41.9 MiB/s 29.0 MiB/s
128k 42.5 MiB/s 39.5 MiB/s 42.1 MiB/s 28.3 MiB/s
256k 42.6 MiB/s 39.9 MiB/s 41.7 MiB/s 28.9 MiB/s
512k 43.8 MiB/s 39.7 MiB/s 43.0 MiB/s 30.9 MiB/s
1M 43.7 MiB/s 40.3 MiB/s 41.7 MiB/s 32.0 MiB/s
Banana Pi SATA disk read results (SSD)
-----------------------------------------------------------
block filesystem type
size -------------------------------------------
(bytes)ext2 ext3 ext4 vfat
8 2.0MiB/s 2.0MiB/s 2.1MiB/s 2.1MiB/s
16 4.2MiB/s 4.2MiB/s 4.2MiB/s 4.2MiB/s
32 9.0MiB/s 8.9MiB/s 9.1MiB/s 8.9MiB/s
64 17.2MiB/s 17.1MiB/s 17.0MiB/s 17.0MiB/s
128 31.4MiB/s 31.6MiB/s 31.5MiB/s 31.1MiB/s
256 53.7MiB/s 54.7MiB/s 55.0MiB/s 55.0MiB/s
512 83.6MiB/s 85.8MiB/s 86.6MiB/s 85.8MiB/s
1k 115.6MiB/s 126.0MiB/s 123.6MiB/s 121.9MiB/s
2k 139.4MiB/s 151.5MiB/s 154.0MiB/s 152.5MiB/s
4k 160.3MiB/s 181.1MiB/s 175.1MiB/s 180.8MiB/s
8k 185.1MiB/s 186.0MiB/s 180.4MiB/s 183.9MiB/s
16k 182.8MiB/s 183.9MiB/s 176.4MiB/s 184.5MiB/s
32k 173.2MiB/s 161.3MiB/s 168.8MiB/s 171.2MiB/s
64k 163.0MiB/s 151.4MiB/s 158.6MiB/s 161.3MiB/s
128k 159.0MiB/s 147.2MiB/s 154.0MiB/s 156.5MiB/s
256k 156.0MiB/s 147.3MiB/s 152.0MiB/s 155.4MiB/s
512k 156.6MiB/s 145.2MiB/s 153.0MiB/s 156.3MiB/s
1M 157.9MiB/s 148.1MiB/s 154.8MiB/s 156.4MiB/s
Since the throughput is more than 150MB/sec, it does confirm that the SATA port in the Banana Pi is SATA 2.0. And I suspect that it could get closer to the SATA 2.0 limit of 300MB/sec if I used a brand new SSD. I have to admit that I am so much more impressed by a Banana Pi especially when compared to a Raspberry Pi. Especially for pushing large amounts of data about fast, it is brilliant hardware.
Subscribe to:
Posts (Atom)



