Marks Rpi Cluster continues to run 24/7, despite a brief network outage yesterday where I had to reboot the router.
You can see when I ran the cluster down to apply updates that required a reboot around 5 days ago with that dip in production.
I assembled the stackable Pi case, put the 4 loose Pi5's into it and used zip ties to hold the 80mm Noctua fan to its side. Airflow is better than the other stackable case so I'll need to swap the other case fan out for a Noctua when I get time.
I also rearranged the support nodes so there is one less with the proxy server now also a storage node (it had a USB SSD anyway). This brings Mark Rpi Cluster to:
24 x Pi4 (8GB) as compute nodes
8 x Pi5 (4GB) as compute nodes
2 x Pi5 (4GB) as support nodes
All the Pi4's are running Einstein@home BRP4 work. The Pi5's are also running Einstein@home but have Asteroids@home as a backup project so when they can't get Einstein work they'll run it. One of the Pi5's picked up some Asteroids work which it has now decided to swap out so it can run Einstein again, hopefully it will finish the Asteroids work soon.
19 July update
The four older Pi5 passed 1 million credits each for Einstein this week. I have put these four into run-off mode in order
to change the fan on its stackable case. It will take a couple of hours to finish their current work.
The newer Pi5's have now got their host average for Einstein up to the older Pi5's which is around 2500 credits. By comparison the oldest Pi4 is on 1.718 million credits but it had a couple of years head-start.
In keeping with the hardware updates, Marks Rpi Cluster is also getting some new kit. I have been buying some Pi5 (4GB) but the resellers are limiting them to one per customer so it takes multiple orders. I'm adding another 4 of them to the cluster. I have ordered one of those stackable things you put them in made from clear acrylic and some metal posts, waiting for it to come from China.
The Pi4's continue processing Einstein BRP4 work. The Pi5's are also doing Einstein BRP4 work but have done a few Asteroids tasks in between. The Pi5's are up to 900M credits for Einstein. It will take them a while to beat the Pi4's which have been running longer.
27th of June update
The stackable Pi Case arrived. All 4 Pi5's arrived and are up and running. The 4 new Pi5's are sitting loose on a table top at the moment. That brings the cluster up to 8 Pi5's for compute. I need to assemble the case which has no instructions so its going to be trial and error.
I had issues with the new Pi5's losing their network connection when plugged into the same switches that the EC12's use. Each EC12 has 12 x Pi4's plugged into a 16 port switch. Each 16 port switch is in turn plugged directly into the router. I dug out an old 5 port gigabit switch, plugged it into a different switch and the 4 new Pi's are connected to it which seems to have resolved the connectivity issue.
I got a couple of Noctua 80mm 5v fans. I'll attach one to the side so it draws air across the Pis. The Pi5's all have the official active cooling fan/heatsink so its there to help them. There isn't much space between each layer. The other case I have is using a slim 80x10mm 5v fan which doesn't move much air. The Noctua fans are 80x25mm and have a higher airflow. The Noctua fans come with a USB type A adapter cable so I can plug them into one of the Pi5's for power.
Currently the 4 x Pi5 nodes are running Einstein.
There was a debian point release so I need to go through each of the Pi4's and update them before I can bring them online.
Purchasing the Sandisk High Endurance micro SD cards (yes I use them for the Pi's primary storage) I can no longer get the 32GB versions. It seems Sandisk have moved to 64GB as the smallest size these days with a price increase.
