25th of April

Marks Rpi Cluster has been running 24/7 since my last post.

The cluster is primarily doing Einstein BRP4 work which takes a little over 3 hours a work unit when running 4 at a time. I occasionally allow Asteroids@home however their work units can take up to 10 hours each so prefer to run the shorter Einstein work units.

Most of the Pi's have their RAC (recent average credit) up to 1200 credits for Einstein with a few of them pushing to 1300 which is about the highest they go for Einstein work.

 

Disable swap
While working on the Chia farm I was looking at disabling swap. I figured I can do the same with the raspberries. 100MB of swap is pretty useless and I have Pi4 8GB as compute nodes so there is no need to have swap enabled on them. It also should extend the micro SD card life by not using swap. The simple (but temporary way) is to type "sudo swapoff -a" at a terminal prompt and its off until you issue a "sudo swapon" or reboot the Pi. Given my Pis run for months between reboots the temporary solution works for me.

As an aside Debian defaults to 977MB for its swap space. Earlier version of Debian would default to the RAM size but it has been capped since Debian 11 (aka Bullseye).

I could configure a larger swap, but I don't run anything that needs much memory. I am not sure why Raspberry Pi OS doesn't default to a larger value, particularly for the 1GB and 2GB models, or disable swap completely.

There are some rather convoluted ways on the internet of disabling swap permanently but I haven't tried any of them yet.

Back again

Marks Rpi Cluster is back to running 24/7. Due to hot weather and my refusal to use air conditioning the farm had a holiday. The farm is currently doing Einstein@home BRP4 work with a few nodes running Asteroids@home. The Asteroids work units have recently been taking up to 10 and a half hours so I don't run many of them.

Recently Raspberry Pi OS got updated to the 6.1 Linux kernel. The 6.1 kernel was declared a long term support version and Debian are looking at their Bookworm release soon. Bookworm is using the 6.1 kernel. I presume this is so the Raspberry Pi foundation can quickly release Raspberry Pi OS based off Debian Bookworm. At the moment Raspberry Pi OS is based off Debian Bullseye.

The cluster is comprised of a pair of BitScope Edge Cluster12's with 12 x Raspberry Pi 4 (8GB) nodes in each providing the compute capability. There are also 3 x Raspberry Pi 4 (2GB) support nodes.

14th of February

Marks Rpi Cluster is back and running after a 3 week weather-induced break. We're back to running Einstein@home BRP4 work on all the compute nodes. Understandably my recent average credit has taken quite a hit with the cluster being off for so long.

The cluster just powered up as normal and I went through each node and applied OS updates and rebooted them before resuming computing.

Einstein have a bunch of new BRP4 work units with "guppi" in the work unit name. We saw these previously at Seti@home. According to a message board post from 2016 it may stand for Greenbank Ultimate Pulsar Processing Instrument. As to when the observations were made we can't tell from their name.

I noticed the Asteroids@home project has come back online. Their server crashed over a year ago and had to get new hardware. It would appear they've been back since November 2022. The work using are used to compute the shape of asteroids using a light-curve method. They have released a ARMv8 app which the Pi's can run (they had a 32 bit app previously). They also have x86 and Cuda apps. I've attached a couple of Pi's to the project and have them computing now. Its looking they take 6 hours per work unit when running 4 at a time and they only use 9MB of memory.

7th of January 2023

After a week of being off the weather cooled down and Marks Rpi Cluster is running again.

During the week of down time I took the opportunity to replace the SD card on the Pi that kept dropping off the network. That meant taking the front panel off the Edge Cluster 12 (EC12) to get to it. While there I noticed it had a lot of dust around the two fans so I gave it a bit of clean with a paint brush and a blower. Having cleaned one EC12 I then had to clean the other one given they've both been running 24/7 since I got them.

I managed to get another Pi4 2GB to use as a proxy server. I'm trying to work out some way of load balancing the proxy servers. If worse comes to worse I can just have each EC12 using a different proxy server but that isn't ideal.