Armbian headless server - power usage

[roel]

Hello,

 

I have a cubieboard A10 connected to a SSD drive (transcend) trough the sata port.  I use it as a headless server to run transmission, nfs and barracudadrive.

The board runs from a sandisk 16gb µSD card, the data is on the SSD (with dynamic links).  I use the standard legacy armbian image without DE.

The SSD is powered by the board and the board is powered with a 2A 5V power supply, comming from a sony psp.

 

If I measure the powerconsumption at the wallsocket, I measure 6W in idle and it goes up to 10W in use (so just enough for the power supply). 

 

Because there is a specialist here in powerconsumption (tkaiser), I wonder if it isn't possible to make a guide (+ tips&tricks) to reduce power usage on headless server's.  This may also include how to chose your HDD/SSD.

 

Another question I have regarding this is the following one:

Can you wake up/shutdown the cubie regarding the ethernet connection?  So if the ethernet goes down, I want the cubie to shutdown or go in standby (to consume as less power possible), if the ethernet goes up again it should wake up again.  Is this possible?

[tkaiser]

Just a few quick notes:

• I would disconnect the SSD and look how consumption 'at the wallsocket' looks like then. If it's still high stop thinking about the board and look for a better PSU instead (an idle A10 board should not exceed 1W on its own or 'measured without PSU')
• I never used Transcend SSDs, when I choose SSDs I look at reviews for performance and power consumption numbers and decide then
• On your Cubieboard it might be possible to read out SSD consumption (that's board specific so no idea whether that will work or not). AXP209 provides a bunch of sysfs nodes to monitor current/voltage, just have a look whether the SSD adds to consumption reported or not -- by removing it entirely and check with hdparm if it reacts to sleep/standby attempts)

AFAIK WOL (wake on LAN) is not possible with these SBC (only dealt with A20 boards with RTL8211E PHY and since the MAC lives inside the SoC it seems it's impossible to get something reacting on the magic packets -- but I might be wrong and also don't care that much since with appropriate settings idle consumption of any Allwinner board is that low that WOL is more or less useless)

[roel]

tkaiser,

 

seems there is a lot enabled when checking with hdparm:

 

-> hdparm -I

/dev/sda:

ATA device, with non-removable media
    Model Number:       TS128GSSD740                            
    Serial Number:      20131128B11844012304
    Firmware Revision:  SVN198  
    Transport:          Serial, ATA8-AST, SATA 1.0a, SATA II Extensions, SATA Rev 2.5, SATA Rev 2.6, SATA Rev 3.0
Standards:
    Supported: 9 8 7 6 5
    Likely used: 9
Configuration:
    Logical        max    current
    cylinders    16383    16383
    heads        16    16
    sectors/track    63    63
    --
    CHS current addressable sectors:   16514064
    LBA    user addressable sectors:  250069680
    LBA48  user addressable sectors:  250069680
    Logical  Sector size:                   512 bytes
    Physical Sector size:                   512 bytes
    Logical Sector-0 offset:                  0 bytes
    device size with M = 1024*1024:      122104 MBytes
    device size with M = 1000*1000:      128035 MBytes (128 GB)
    cache/buffer size  = unknown
    Form Factor: less than 1.8 inch
    Nominal Media Rotation Rate: Solid State Device
Capabilities:
    LBA, IORDY(can be disabled)
    Queue depth: 32
    Standby timer values: spec'd by Standard, no device specific minimum
    R/W multiple sector transfer: Max = 16    Current = 16
    Advanced power management level: 254
    DMA: mdma0 mdma1 mdma2 udma0 udma1 udma2 udma3 udma4 *udma5
         Cycle time: min=120ns recommended=120ns
    PIO: pio0 pio1 pio2 pio3 pio4
         Cycle time: no flow control=120ns  IORDY flow control=120ns
Commands/features:
    Enabled    Supported:
       *    SMART feature set
            Security Mode feature set
       *    Power Management feature set
       *    Write cache
       *    Look-ahead
       *    Host Protected Area feature set
       *    WRITE_BUFFER command
       *    READ_BUFFER command
       *    NOP cmd
       *    DOWNLOAD_MICROCODE
       *    Advanced Power Management feature set
            SET_MAX security extension
       *    48-bit Address feature set
       *    Device Configuration Overlay feature set
       *    Mandatory FLUSH_CACHE
       *    FLUSH_CACHE_EXT
       *    SMART error logging
       *    SMART self-test
       *    General Purpose Logging feature set
       *    WRITE_{DMA|MULTIPLE}_FUA_EXT
       *    64-bit World wide name
       *    WRITE_UNCORRECTABLE_EXT command
       *    {READ,WRITE}_DMA_EXT_GPL commands
       *    Segmented DOWNLOAD_MICROCODE
       *    Gen1 signaling speed (1.5Gb/s)
       *    Gen2 signaling speed (3.0Gb/s)
       *    Gen3 signaling speed (6.0Gb/s)
       *    Native Command Queueing (NCQ)
       *    Host-initiated interface power management
       *    Phy event counters
       *    unknown 76[13]
       *    unknown 76[14]
       *    unknown 76[15]
            DMA Setup Auto-Activate optimization
            Device-initiated interface power management
       *    Software settings preservation
            unknown 78[8]
       *    SMART Command Transport (SCT) feature set
       *    SCT Error Recovery Control (AC3)
       *    SCT Features Control (AC4)
       *    SCT Data Tables (AC5)
       *    Data Set Management TRIM supported (limit 2 blocks)
Security:
    Master password revision code = 65534
        supported
    not    enabled
    not    locked
    not    frozen
    not    expired: security count
        supported: enhanced erase
    2min for SECURITY ERASE UNIT. 2min for ENHANCED SECURITY ERASE UNIT.
Logical Unit WWN Device Identifier: 5000000000000000
    NAA        : 5
    IEEE OUI    : 000000
    Unique ID    : 000000000
Checksum: correct

 

-> hdparm -B

/dev/sda:
 APM_level    = 254 

 

This one can probably set lower.

 

hdparm -y and hdparm -Y doesn't change anything on the powerconsumption.  If I disconnect the drive the powerconsumption settles on 3,5W.

 

I tried another PSU but the power consumption stays the same.  I have only from these switching power supplies that are included with a device.

I didn't bought the SSD, it's one I took out of a PC they would throw away.  So in that assumption it was cheap and therefore may use some more power.

 

I have another SSD from sandisk.  I will try that, maybe it uses less power.


 

[roel]

I did some tests today with the power supply.

 

First I tried with a apple ipad PSU (should be quality) and the usb power cable included with the cubie.  The board wouldn't even boot.

Then I tried with another usb-cable a received together with a solar panel they gave away for free.  A miracle, the consumption got down to 3,3W with the transcend ssd, when I tried the sandisk, consumption was 3,1W, without it settled at 1,5W.

 

Then I tried with another usb psu I ones bought in a local hardware shop.  With the transcend disk it was 2,1W, with the sandisk 2,5W.  Strange, now the sandisk used more. 

 

I also checked with the psu of my mele A1000.  Not so good results: 6W.

 

I checked the cables of the psu's and the usb-cables.  On the Sony is mentioned 20AWG, not so bad though.  The others don't mention anything, only black cables.

 

Then I tried again with the sony psu, now it settles at 2,5W.  That's something really different then the days before when I measured 6W idle.  Probably some bad contact as it was a psu I didn't use already for a long time.  Probably with changing the cables while testing the connector cleaned itself.

 

Conclusion:  it's worth investigating those psu and usb cables.

[tkaiser]

For my understanding: you used USB-Type-A to 4.0/1.7mm cables with the USB chargers? Then I had an incident a few days ago where I powered down an OPi Plus 2E and realized the next day that the connected SSD was pretty warm and consumption according to powermeter above 3W, I rebooted and temperature reported via SMART was above 40°C -- so the SSD was in a somewhat strange state for hours (Samsung EVO750 -- not that much experiences with these now)

 

BTW: since Apple peripherals are rather expensive there exist a lot of counterfeit chargers, just as one example: http://www.righto.com/2012/10/a-dozen-usb-chargers-in-lab-apple-is.html(fun with the oscilloscope).