Transient missing spots events may require WD server CPU upgrade


Rob Robinett
 

This kind of problem has been reported by others in the Wednesday WD Zoom call and fixed by upgrading to a more powerful CPU than the ARM on the Pi4.  It appears that with the summer months and increasing sunspot cycle, the Pi CPUs can't keep up decoding the number of spots in the wav file captures.  However it is hard to observe those 'out of CPU' events.
 
Most of the top spotting sites have migrated from Pi4s to i86 servers.  An inexpensive, low power and high performance option is the Lenovo M93P Tiny PC (don't get the larger model) with i5 CPU which can be purchased used or refurbished for $100 or less on ebay:  https://tinyurl.com/emrjnw  The M93P consumes 12W peaking at 36W while decoding spots.
 
WD requires very little RAM space, so a 2 or 4GB machine would be fine.  I bought mine without HD and added a SSD and installed Ubuntu 20.04 LTS. But since WD does very little disk  IO to the root file system, using a rotating HD might be a better choice for longer life.
 


Edward Hammond
 

I noticed last year that my decode cycles were pushing the Pi's limits.  Your mileage may vary, and it certainly will void any warranty, but overclocking the Pi4 is also a possibility.

I experimented with that option and, on my example at least, I could dial in a ~12-14% clock speed boost stably (IIRC).  This did shorten the decoding cycle.

Any higher clock rate, again on my particular Pi, and it started behaving erratically.

Edward

On 5/25/21 12:37 PM, Rob Robinett wrote:
This kind of problem has been reported by others in the Wednesday WD Zoom call and fixed by upgrading to a more powerful CPU than the ARM on the Pi4.  It appears that with the summer months and increasing sunspot cycle, the Pi CPUs can't keep up decoding the number of spots in the wav file captures.  However it is hard to observe those 'out of CPU' events.
Most of the top spotting sites have migrated from Pi4s to i86 servers.  An inexpensive, low power and high performance option is the Lenovo M93P Tiny PC (don't get the larger model) with i5 CPU which can be purchased used or refurbished for $100 or less on ebay: https://tinyurl.com/emrjnw  The M93P consumes 12W peaking at 36W while decoding spots.
WD requires very little RAM space, so a 2 or 4GB machine would be fine.  I bought mine without HD and added a SSD and installed Ubuntu 20.04 LTS. But since WD does very little disk  IO to the root file system, using a rotating HD might be a better choice for longer life.


Edward Hammond
 

Sorry, a correction.  It was FT8 that was pushing my Pi to its limits then, not WSPR.  But I would think overclocking would similarly help in this situation.

EH

On 5/25/21 3:29 PM, Edward Hammond wrote:

I noticed last year that my decode cycles were pushing the Pi's limits.  Your mileage may vary, and it certainly will void any warranty, but overclocking the Pi4 is also a possibility.

I experimented with that option and, on my example at least, I could dial in a ~12-14% clock speed boost stably (IIRC).  This did shorten the decoding cycle.

Any higher clock rate, again on my particular Pi, and it started behaving erratically.

Edward



On 5/25/21 12:37 PM, Rob Robinett wrote:
This kind of problem has been reported by others in the Wednesday WD Zoom call and fixed by upgrading to a more powerful CPU than the ARM on the Pi4.  It appears that with the summer months and increasing sunspot cycle, the Pi CPUs can't keep up decoding the number of spots in the wav file captures.  However it is hard to observe those 'out of CPU' events.
Most of the top spotting sites have migrated from Pi4s to i86 servers.  An inexpensive, low power and high performance option is the Lenovo M93P Tiny PC (don't get the larger model) with i5 CPU which can be purchased used or refurbished for $100 or less on ebay: https://tinyurl.com/emrjnw  The M93P consumes 12W peaking at 36W while decoding spots.
WD requires very little RAM space, so a 2 or 4GB machine would be fine.  I bought mine without HD and added a SSD and installed Ubuntu 20.04 LTS. But since WD does very little disk  IO to the root file system, using a rotating HD might be a better choice for longer life.


Rob Robinett
 

In WD V3.0 I will be adding support for FST4W 120/300/900/1800 and using those modes, especially the 900/1800, will surely require more CPU than any Pi can offer.

On Tue, May 25, 2021 at 1:06 PM Edward Hammond <manager@...> wrote:
Sorry, a correction.  It was FT8 that was pushing my Pi to its limits
then, not WSPR.  But I would think overclocking would similarly help in
this situation.

EH


On 5/25/21 3:29 PM, Edward Hammond wrote:
>
> I noticed last year that my decode cycles were pushing the Pi's
> limits.  Your mileage may vary, and it certainly will void any
> warranty, but overclocking the Pi4 is also a possibility.
>
> I experimented with that option and, on my example at least, I could
> dial in a ~12-14% clock speed boost stably (IIRC).  This did shorten
> the decoding cycle.
>
> Any higher clock rate, again on my particular Pi, and it started
> behaving erratically.
>
> Edward
>
>
>
> On 5/25/21 12:37 PM, Rob Robinett wrote:
>> This kind of problem has been reported by others in the Wednesday WD
>> Zoom call and fixed by upgrading to a more powerful CPU than the ARM
>> on the Pi4.  It appears that with the summer months and increasing
>> sunspot cycle, the Pi CPUs can't keep up decoding the number of spots
>> in the wav file captures.  However it is hard to observe those 'out
>> of CPU' events.
>> Most of the top spotting sites have migrated from Pi4s to i86
>> servers.  An inexpensive, low power and high performance option is
>> the Lenovo M93P Tiny PC (don't get the larger model) with i5 CPU
>> which can be purchased used or refurbished for $100 or less on ebay:
>> https://tinyurl.com/emrjnw  The M93P consumes 12W peaking at 36W
>> while decoding spots.
>> WD requires very little RAM space, so a 2 or 4GB machine would be
>> fine.  I bought mine without HD and added a SSD and installed Ubuntu
>> 20.04 LTS. But since WD does very little disk  IO to the root file
>> system, using a rotating HD might be a better choice for longer life.
>>
>
>








--
Rob Robinett
AI6VN
mobile: +1 650 218 8896


Edward Hammond
 

Okay.  I guess it's time to suck it up and install Ubuntu on that old Intel NUC!

EH

On 5/25/21 4:10 PM, Rob Robinett wrote:
In WD V3.0 I will be adding support for FST4W 120/300/900/1800 and using those modes, especially the 900/1800, will surely require more CPU than any Pi can offer.

On Tue, May 25, 2021 at 1:06 PM Edward Hammond <manager@muskrat.farm> wrote:

Sorry, a correction.  It was FT8 that was pushing my Pi to its limits
then, not WSPR.  But I would think overclocking would similarly
help in
this situation.

EH


On 5/25/21 3:29 PM, Edward Hammond wrote:
>
> I noticed last year that my decode cycles were pushing the Pi's
> limits.  Your mileage may vary, and it certainly will void any
> warranty, but overclocking the Pi4 is also a possibility.
>
> I experimented with that option and, on my example at least, I
could
> dial in a ~12-14% clock speed boost stably (IIRC).  This did
shorten
> the decoding cycle.
>
> Any higher clock rate, again on my particular Pi, and it started
> behaving erratically.
>
> Edward
>
>
>
> On 5/25/21 12:37 PM, Rob Robinett wrote:
>> This kind of problem has been reported by others in the
Wednesday WD
>> Zoom call and fixed by upgrading to a more powerful CPU than
the ARM
>> on the Pi4.  It appears that with the summer months and increasing
>> sunspot cycle, the Pi CPUs can't keep up decoding the number of
spots
>> in the wav file captures.  However it is hard to observe those
'out
>> of CPU' events.
>> Most of the top spotting sites have migrated from Pi4s to i86
>> servers.  An inexpensive, low power and high performance option is
>> the Lenovo M93P Tiny PC (don't get the larger model) with i5 CPU
>> which can be purchased used or refurbished for $100 or less on
ebay:
>> https://tinyurl.com/emrjnw <https://tinyurl.com/emrjnw> The
M93P consumes 12W peaking at 36W
>> while decoding spots.
>> WD requires very little RAM space, so a 2 or 4GB machine would be
>> fine.  I bought mine without HD and added a SSD and installed
Ubuntu
>> 20.04 LTS. But since WD does very little disk  IO to the root file
>> system, using a rotating HD might be a better choice for longer
life.
>>
>
>








--
Rob Robinett
AI6VN
rob@robinett.us <mailto:rob@robinett.us>
mobile: +1 650 218 8896


Jim Lill
 

The NUC may not be enough...    I use a single Odroid XU4 to handle 26 channels worth of WD and haven't seen issues yet (but you look more carefully now)

On 5/25/21 5:02 PM, Edward Hammond wrote:
Okay. I guess it's time to suck it up and install Ubuntu on that old Intel NUC!

EH


On 5/25/21 4:10 PM, Rob Robinett wrote:
In WD V3.0 I will be adding support for FST4W 120/300/900/1800 and using those modes, especially the 900/1800, will surely require more CPU than any Pi can offer.

On Tue, May 25, 2021 at 1:06 PM Edward Hammond <manager@muskrat.farm> wrote:

    Sorry, a correction.  It was FT8 that was pushing my Pi to its limits
    then, not WSPR.  But I would think overclocking would similarly
    help in
    this situation.

    EH


    On 5/25/21 3:29 PM, Edward Hammond wrote:
    >
    > I noticed last year that my decode cycles were pushing the Pi's
    > limits.  Your mileage may vary, and it certainly will void any
    > warranty, but overclocking the Pi4 is also a possibility.
    >
    > I experimented with that option and, on my example at least, I
    could
    > dial in a ~12-14% clock speed boost stably (IIRC). This did
    shorten
    > the decoding cycle.
    >
    > Any higher clock rate, again on my particular Pi, and it started
    > behaving erratically.
    >
    > Edward
    >
    >
    >
    > On 5/25/21 12:37 PM, Rob Robinett wrote:
    >> This kind of problem has been reported by others in the
    Wednesday WD
    >> Zoom call and fixed by upgrading to a more powerful CPU than
    the ARM
    >> on the Pi4.  It appears that with the summer months and increasing
    >> sunspot cycle, the Pi CPUs can't keep up decoding the number of
    spots
    >> in the wav file captures.  However it is hard to observe those
    'out
    >> of CPU' events.
    >> Most of the top spotting sites have migrated from Pi4s to i86
    >> servers.  An inexpensive, low power and high performance option is
    >> the Lenovo M93P Tiny PC (don't get the larger model) with i5 CPU
    >> which can be purchased used or refurbished for $100 or less on
    ebay:
    >> https://tinyurl.com/emrjnw <https://tinyurl.com/emrjnw> The
    M93P consumes 12W peaking at 36W
    >> while decoding spots.
    >> WD requires very little RAM space, so a 2 or 4GB machine would be
    >> fine.  I bought mine without HD and added a SSD and installed
    Ubuntu
    >> 20.04 LTS. But since WD does very little disk  IO to the root file
    >> system, using a rotating HD might be a better choice for longer
    life.
    >>
    >
    >








--
Rob Robinett
AI6VN
rob@robinett.us <mailto:rob@robinett.us>
mobile: +1 650 218 8896