Hi volunteers,

I am trying to get Asteroids@home running on my Pi Kubernetes Cluster. The Pis are running Raspberry Pi OS (64 bit)
Project response "Message from server: This project doesn't support computers of type aarch64-unknown-linux-gnu"

BOINC Client starts up like that

Starting BOINC client version 7.16.11 for aarch64-unknown-linux-gnu
log flags: file_xfer, sched_ops, task
Libraries: libcurl/7.68.0 OpenSSL/1.1.1f zlib/1.2.11 brotli/1.0.9 libidn2/2.3.0 libpsl/0.21.0 (+libidn2/2.3.0) libssh/0.9.3/openssl/zlib nghttp2/1.41.0 librtmp/2.3
Data directory: /opt/boinc_taskstorage
No usable GPUs found
libc: Ubuntu GLIBC 2.32-0ubuntu3 version 2.32
Host name: boinc
Processor: 4 ARM BCM2835 [Impl 0x41 Arch 8 Variant 0x0 Part 0xd08 Rev 3]
Processor features: fp asimd evtstrm crc32 cpuid
OS: Linux Ubuntu: Ubuntu 20.10 [5.10.17-v8+|libc 2.32 (Ubuntu GLIBC 2.32-0ubuntu3)]
Memory: 7.68 GB physical, 0 bytes virtual
Disk: 28.41 GB total, 15.34 GB free
Local time is UTC +1 hours

I tried with alternative platforms in a custom cc_config:
<alt_platform>arm-unknown-linux-gnueabihf</alt_platform>

but this leads to computation errors:

Name ps_210311_input_55539_22_0
Application Period Search Application 102.13
Workunit name ps_210311_input_55539_22
State Computation error
Received 3/12/2021 8:40:41 PM
Report deadline 3/23/2021 8:40:40 AM
Estimated app speed 11,91 GFLOPs/sec
Estimated task size 1.380.023 GFLOPs
CPU time at last checkpoint 00:00:00
CPU time 00:00:00
Elapsed time 00:00:00
Estimated time remaining 00:00:00
Fraction done 100%
Virtual memory size 0,00 MB
Working set size 0,00 MB

Any suggestions how to tweak BOINC with an alternative_platform or
is it more likely that Kyong has to recompile something ?

Btw Rosetta, LHC and Einstein are already running well on this cluster.

Cheers

Jan

Hi volunteers,

I think I found a solution for my issue.

As stated above I am running BOINC Kubernetes (K3S) cluster in Raspberry Pi 4 models. I build the container image for the pods by myself based on the original boinc/client buildcode on github.

The container image I build had only 64 Bit libraries. Thats why Astroids@Home 32 Bit Application generated "Computation errors". I rebuild the container image and added the needed 32Bit architecture armhf and also the libstdc++6:armhf library. Docker buildfile looks like this now:

FROM arm64v8/ubuntu:rolling

LABEL maintainer="BOINC" \
      description="A lightweight BOINC client on ARMv8 64-bit architecture."

# Global environment settings
ENV BOINC_GUI_RPC_PASSWORD="123" \
    BOINC_REMOTE_HOST="" \
    BOINC_CMD_LINE_OPTIONS="" \
    DEBIAN_FRONTEND=noninteractive

# Copy files
COPY bin/ /usr/bin/

# Configure
WORKDIR /var/lib/boinc

# BOINC RPC port
EXPOSE 12345

# Install (also 32Bit architecture and libraries)
RUN dpkg --add-architecture armhf && apt-get update && apt-get -y upgrade -qq && apt-get install libstdc++6:armhf -y && apt-get install -y --no-install-recommends \
# Install Time Zone Database
        tzdata \
# Install BOINC Client and some tools
    boinc-client tar sed wget && \

# Cleaning up
    apt-get autoremove -y && \
    rm -rf /var/lib/apt/lists/*

CMD ["start-boinc.sh"]

The zipped containerimage grew from ~120MB to 140MB but thats because of the needed 32Bit libraries.

I also added two alternative platforms to the cc_config.xml

<cc_config>
  <options>
  <alt_platform>arm-unknown-linux-gnueabihf</alt_platform>
  <alt_platform>armv7l-unknown-linux-gnueabihf</alt_platform>
  </options>
</cc_config>

Now Asteroids@Home tasks are fetched and processed (btw. also Universe@Home). Yet without any "Computation errors". Looks like each Asteroids@Home task needs about 15h to finish.

I wonder if a task could be finished faster if there would be a 64Bit Asteroids@Home application for arch64-unknown-linux-gnu platform.

Happy Crunching

Jan

The current state of apps for the Rpi is summarised in this post from August 2020.

1. The ARMv6 Pi’s all fail (that’s Pi zero and the original Pi).
2. The current app works on Pi2 and up but doesn’t use neon which makes it extremely slow.
3. There isn’t any support for aarch64

The current state of apps for the Rpi is summarised in this post from August 2020.

1. The ARMv6 Pi’s all fail (that’s Pi zero and the original Pi).
2. The current app works on Pi2 and up but doesn’t use neon which makes it extremely slow.
3. There isn’t any support for aarch64

Has this changed since your post in 2021?

a native aarch64 Linux app is available https://asteroidsathome.net/boinc/apps.php

32-bit Linux ARM apps (such as the RPi 3) should still run, but they’re very slow and may struggle to finish even a single work unit before the deadline.

The current state of apps for the Rpi is summarised in this post from August 2020.

1. The ARMv6 Pi’s all fail (that’s Pi zero and the original Pi).
2. The current app works on Pi2 and up but doesn’t use neon which makes it extremely slow.
3. There isn’t any support for aarch64

Has this changed since your post in 2021?

Yes it has changed since then. The project now has aarch64 support for the Pi4's and Pi5's.

The latest app is the 102.21 Period Search app (aarch64_armv8a_asimd) which I run on Pi5's without issue and it should also run on Pi4's but they'll be slower. On a Pi5 tasks take around 3 hours and 10-15 minutes.

---

MarksRpiCluster

a native aarch64 Linux app is available https://asteroidsathome.net/boinc/apps.php

32-bit Linux ARM apps (such as the RPi 3) should still run, but they’re very slow and may struggle to finish even a single work unit before the deadline.

I fired up a couple of Pi3's and a Pi4.

Aarch64 (64 bit) can run on later model Pi3's because they use the same SoC as the Pi4's https://asteroidsathome.net/boinc/results.php?hostid=833482. This ran the armv8a app and was around 110k seconds, it also ran two Einstein BRP4 tasks at the same time.

Armhf (32 bit) which took 144k seconds but still managed to complete using the 102.20 app. It ran a single Einstein BRP4 at the same time https://asteroidsathome.net/boinc/results.php?hostid=833510. An armv7l app with neon might speed things up but I am not sure if it justifies the development effort.

The Pi4 averaged 33460 seconds running 4 at a time. It used the armv8a app https://asteroidsathome.net/boinc/results.php?hostid=833481

---

MarksRpiCluster

I fired up a couple of Pi3's and a Pi4.

Aarch64 (64 bit) can run on later model Pi3's because they use the same SoC as the Pi4's https://asteroidsathome.net/boinc/results.php?hostid=833482. This ran the armv8a app and was around 110k seconds, it also ran two Einstein BRP4 tasks at the same time.

Armhf (32 bit) which took 144k seconds but still managed to complete using the 102.20 app. It ran a single Einstein BRP4 at the same time https://asteroidsathome.net/boinc/results.php?hostid=833510. An armv7l app with neon might speed things up but I am not sure if it justifies the development effort.

I downgraded the aarch64 Pi3 to 32 bit and got it running the 102.20 arm app. I have had both Pi3's running 4 x tasks. Average elapsed time is 151k seconds or approx 43 hours for them to complete.

---

MarksRpiCluster