To my pleaseant surprise the software I used for my Pentium Pro FreeBSD cluster is still available as source code. It is great for making all your mistakes global. :-/
I finally managed to attach the project to my Pi 3. Thanks poppageek !
But I noticed a problem. The logfile stderrdae.txt is getting bigger and bigger. In the meantime its around 8 MB, the stderrdae.old nearly 22 MB.
It contains mostly one single message, but some 1000 times :
dir_open: Could not open directory '/dev/input/mice' from '/var/lib/boinc-client'
As I did just a normal installation using the March Version of Rasbian I think that more people could have this issue, but just do not notice.
I guess this behaviour is not directly related to Einstein but a BOINC issue.
I finally managed to attach the project to my Pi 3. Thanks poppageek !
But I noticed a problem. The logfile stderrdae.txt is getting bigger and bigger. In the meantime its around 8 MB, the stderrdae.old nearly 22 MB.
It contains mostly one single message, but some 1000 times :
dir_open: Could not open directory '/dev/input/mice' from '/var/lib/boinc-client'
As I did just a normal installation using the March Version of Rasbian I think that more people could have this issue, but just do not notice.
I guess this behaviour is not directly related to Einstein but a BOINC issue.
Any ideas ?
I too noticed this when trying to figure out why I could not connect to server. Forgot about it. I will look into it as I agree it could cause problems. Thanks.
Cheers!
EDIT:
It appears that messages are written to stderrdae.txt until it reaches a certain size, from the looks of them on mine about 21megs. The file is renamed stderrdae.old and a new stderrdae.txt is created and written to. Repeat. So it looks like the size is limited. I will keep an eye on it but I do not think it's a problem.
If 2x 21MB is an issue or the size is larger, you have at least two options:
You can either add the user boinc to system group input or purge those messages from the log through awk..
This will need gawk (1.8MB installed), then put following to your crontab:
0,30 * * * * gawk -i inplace '!/dir_open/ { print }' /var/lib/boinc-client/stderrdae.txt
Most likely you can customize the log size as well, but hey, it's already worked around, I'm lazy...
Hello guys.
This is an awesome thread, I've been following it for a while, although I didn't read 100% of all messages.
I think you have gathered some really useful statistics on the Pi, so I believe this is the proper place to ask this:
I want run BOINC on a Pi3 with a focus on efficiency. The philosophy I want to follow is that I can get more results done if I spend my resources on getting more Pis instead of spending in energy and coolers. That means more Pis instead of overclocked Pis with a fans.
Do you have any data to back up this idea? I think RAC/W would be enough, I could add infra costs later. It would be nice to have RAC/W for the Pi3 on fan, sink and nothing attached (maybe needs to be underclocked?).
I want run BOINC on a Pi3 with a focus on efficiency. The philosophy I want to follow is that I can get more results done if I spend my resources on getting more Pis instead of spending in energy and coolers. That means more Pis instead of overclocked Pis with a fans.
Do you have any data to back up this idea? I think RAC/W would be enough, I could add infra costs later. It would be nice to have RAC/W for the Pi3 on fan, sink and nothing attached (maybe needs to be underclocked?).
Nice idea to look at this from a RAC/W (or in other words: credits / kWh) perspective.
Some random thoughts:
It could be that the RPI2 is more efficient than the RPi3, actually. The RPi2 runs happily without active cooling under full BOINC load.
Underclocking: The Raspi 3 will automatically underclock itself when reaching about 80deg C. When this happens will depend on external factors like airflow over the CPU (in a case/exposed) and of course ambient temperature. So it's a bit difficult to come up with numbers that do not take these factors into account.
Its also not trivial to measure the power drawn by a Raspi, most of the Kill-a-Watt-style meters are designed for higher loads and will not give reliable readings for just a single Pi (and wall socket power consumption is influenced by PSU efficiency anyway).
The deed is done. Pi Team ID
)
The deed is done. Pi Team ID is PG3.14. ;-)
To my pleaseant surprise the software I used for my Pentium Pro FreeBSD cluster is still available as source code. It is great for making all your mistakes global. :-/
http://www.garbled.net/clusterit.html
I finally managed to attach
)
I finally managed to attach the project to my Pi 3. Thanks poppageek !
But I noticed a problem. The logfile stderrdae.txt is getting bigger and bigger. In the meantime its around 8 MB, the stderrdae.old nearly 22 MB.
It contains mostly one single message, but some 1000 times :
dir_open: Could not open directory '/dev/input/mice' from '/var/lib/boinc-client'
As I did just a normal installation using the March Version of Rasbian I think that more people could have this issue, but just do not notice.
I guess this behaviour is not directly related to Einstein but a BOINC issue.
Any ideas ?
RE: I finally managed to
)
I too noticed this when trying to figure out why I could not connect to server. Forgot about it. I will look into it as I agree it could cause problems. Thanks.
Cheers!
EDIT:
It appears that messages are written to stderrdae.txt until it reaches a certain size, from the looks of them on mine about 21megs. The file is renamed stderrdae.old and a new stderrdae.txt is created and written to. Repeat. So it looks like the size is limited. I will keep an eye on it but I do not think it's a problem.
Cheers!
If 2x 21MB is an issue or the
)
If 2x 21MB is an issue or the size is larger, you have at least two options:
You can either add the user boinc to system group input or purge those messages from the log through awk..
This will need gawk (1.8MB installed), then put following to your crontab:
0,30 * * * * gawk -i inplace '!/dir_open/ { print }' /var/lib/boinc-client/stderrdae.txt
Most likely you can customize the log size as well, but hey, it's already worked around, I'm lazy...
Hello guys. This is an
)
Hello guys.
This is an awesome thread, I've been following it for a while, although I didn't read 100% of all messages.
I think you have gathered some really useful statistics on the Pi, so I believe this is the proper place to ask this:
I want run BOINC on a Pi3 with a focus on efficiency. The philosophy I want to follow is that I can get more results done if I spend my resources on getting more Pis instead of spending in energy and coolers. That means more Pis instead of overclocked Pis with a fans.
Do you have any data to back up this idea? I think RAC/W would be enough, I could add infra costs later. It would be nice to have RAC/W for the Pi3 on fan, sink and nothing attached (maybe needs to be underclocked?).
Thanks
Hi Roberto I'll try and
)
Hi Roberto
I'll try and get ya some numbers in the next couple of days.
Cheers!
RE: I want run BOINC on a
)
Nice idea to look at this from a RAC/W (or in other words: credits / kWh) perspective.
Some random thoughts:
It could be that the RPI2 is more efficient than the RPi3, actually. The RPi2 runs happily without active cooling under full BOINC load.
Underclocking: The Raspi 3 will automatically underclock itself when reaching about 80deg C. When this happens will depend on external factors like airflow over the CPU (in a case/exposed) and of course ambient temperature. So it's a bit difficult to come up with numbers that do not take these factors into account.
Its also not trivial to measure the power drawn by a Raspi, most of the Kill-a-Watt-style meters are designed for higher loads and will not give reliable readings for just a single Pi (and wall socket power consumption is influenced by PSU efficiency anyway).
(Tested) Raspberry Pi 3 vs
)
(Tested) Raspberry Pi 3 vs Raspberry Pi 2: CPU and GPU Benchmarks (+ Burn-in Test)
http://www.geeks3d.com/20160322/tested-raspberry-pi-3-vs-raspberry-pi-2-cpu-and-gpu-benchmarks-burn-in-test/
Includes thermal imaging of Pi3 with and with out heat sink while running sysbench.
As Bikeman said a $15
)
As Bikeman said a $15 Kill-a-watt is not known for accuracy. Just an idea of what is going on.
RPi2 @900 mz 4 NEON_Beta tasks. Copper heat sink no fan on case.
4 watts .06 amps
vcgencmd = 57.8c - 58.9c after running min 10 minutes startup warm
Average of 18 samples
Run = 59,286 seconds
CPU = 59,154 seconds
RPi3 @1200mz 4 NEON_Beta tasks. Copper heat sink and fan plugged into CPIO. Open stack, no case.
with fan: 6 watts .08 amps.
Without fan: 5 watts .06 amps
vcgencmd 56.9c - 58c running min 10 minutes startup warm
Average of 20 samples
Run = 42,494
CPU = 42,420
Enzotech RAM-Kühler BMR-C1
)
Enzotech RAM-Kühler BMR-C1 High Profile - passive > 50 available in stock
http://www.aquatuning.us/air-cooling/passive-coolers/7045/enzotech-ram-kuehler-bmr-c1-high-profile-passiv?sPartner=googleshoppingusa
Cheers!