The cluster is continuing to run 24/7. The Pi3's are doing Einstein BRP4 work and the Pi4's are doing a mix of Einstein and Rosetta work.
Previously I had the file server running the GPS as well as serving files but decided for performance reasons to put the GPS onto its own Pi. The cluster now has 3 support nodes. One is an internet proxy server, two is a file server, and three is the time server with a GPS attached. The reason for having a GPS is to have correct time even if there is no internet connectivity.
Debian did a 10.8 point release today so there were a number of updates for that. In addition the Raspberry Pi foundation updated to the 5.10.11 Linux kernel. That meant I spent a bit of time updating all the Pis in the cluster.
3 comments:
Hey Mark, just found your blog. Great work :)
Have you published processing times per task for Einstein@Home from your Pi4 cluster nodes somewhere on your blog?
I started a Kubernetes based Raspi BOINC cluster in November last year and I am wondering if there are any differences between a native boinc client compared to a containerized boinc-client.
Cheers
Jan
The project supplied app was taking 4+ hours on the Pi4 and 12+ hours on the Pi3.
I am using the optimized Einstein BRP4 app which uses a bit more memory but is a lot quicker. Its only available for aarch64.
The Pi4's are taking 2 hours 22 mins. That is running 2 Einstein and 2 Rosetta at the same time. The Pi3's are taking 4 hours 40 mins. They're only running 3 at a time due to lack of memory.
Awesome, thanks for your feedback. Your numbers show that there is no impact on crunching with K3S vs. Raspian when using the BRP4 app. Because of the CPU reserved for K3S i actually can only run 3 tasks in parallel. Need still some time to fiddle that out ... so happy crunching!
Cheers
Jan
Post a Comment