14th of July

Marks Rpi Cluster continues to run 24/7. As usual it is concentrating on the Einstein@home BOINC project running their BRP4 work units (Binary Radio Pulsar search #4). The Pi5's have managed to get their RAC (recent average credit) up to 1800 while the Pi4's are in the 1200's, which suggests the Pi5 is around 1/3rd faster than a Pi4.

Einstein had a scheduled outage last week where they were rearranging some back-end servers, so I increased the cache on all of them so they keep more work ready to start than usual. I also allow them to run another project if they can't get any work from Einstein. I didn't notice anything unusual when checking on them in the evening and there were no work units from the back-up project.

I noticed a news article on Phoronix and Geoff Geerling also made a video around a patche that can improve the Raspberry Pi5's speed by using NUMA emulation mode. The Linux kernel patch was rejected by the kernel developers in its current form so it needs to be reworked but that doesn't stop the Raspberry Pi foundation from applying it to their version of the kernel. There was also another patch that can reduce idle power consumption. 

Phoronix article for idle power: Suspend-To-Idle-Raspberry-Pi

Phoronix article for NUMA emulation: ARM64-NUMA-Emulation-RPi5 

You can see Geoff's YouTube video HERE.

Added some Pi5's part 2

The Pi case arrived and has now been assembled.

There isn't much room from one Pi to the layer above and the Pi's were reporting temps in the 60's (degrees C). I added a fan on the side to help move the warm air from each layer which has reduced the temperature to the mid to high 50 degrees.

I still need to get short network cables to help tidy things up.

The Pi5's have now passed the RAC (Recent Average Credit) that the Pi4's were getting so they are being more productive.

Added some Pi5's

Marks Rpi Cluster continues to run 24/7.

Seeing as the Edge Cluster from BitScope isn't available for the Pi5 I purchased four Pi5's and associated bits to add them to the cluster.

4 x Raspberry Pi5 (4GB)
4 x Raspberry Pi Active Cooler
4 x Raspberry Pi5 USB-C Power Supply
5 x Sandisk 32GB High Endurance micro SD cards
1 x Power board with 8 outlets

I ordered one of the stackable acrylic cases from eBay as I couldn't find them available from the usual Rpi shops. They are just sheets of clear acrylic and a bunch of metal stand-offs. While I wait for the case I have them sitting loose on an old TV table. I need to buy some short network cables as I used a bunch of 2 meter ones and its like spaghetti junction at the moment. I had to get a power board (aka a power strip) due to all the power supplies. I had a spare 8 port switch which I used and the network cables.

You can see the four Pi5's just sitting on the table with their active coolers. Thats an 8 port switch on the top right with a Pi4 sitting on top of it. Is got an external HDD plugged in. On the bottom left are two more Pi4's. The one with the blue thing on top is a proxy server/Pihole machine and not part of the cluster. The blue thing is a Samsung T5 portable SSD. The Pi in the bottom left corner is a time server. The beige USB cable is plugged into a GPS.

Hopefully the stackable case will arrive soon so I can tidy things up. My aim was to get them going as quickly as possible. The delivery date on the case is supposedly the 29th of June.

The Pi5's are running Asteroids and Einstein work. They have shaved a few hours off for the Asteroids Period Search app. Einstein BRP4 on a Pi4 typically take 4 hours 40 minutes and on the Pi5 seem to be 2 hours 52 minutes (running 4 at a time).

The cluster now consists of:

4 x Pi5 (4GB) compute nodes.
24 x Pi4 (8GB) compute nodes.
Plus a Pi5 and two Pi4 as support nodes.

8th of June

Marks Rpi Cluster continues to run 24/7. Its has been running Einstein BRP4 work for the last couple of months.

Last weekend the project had an issue and all file transfers were suspended. According to their message boards they disabled the transitioner. When I checked on the farm I noticed lots of file transfers stuck and only the remaining work finishing off. At that point after retrying a few transfers, which were still stuck, I decided to let the farm run Asteroids work. Come the Monday and the file transfers had cleared up and we're back to running BRP4.

This week one of the Pi4's locked up. In order to reboot I have to run off all work and power cycle the edge cluster. So I suspend any work waiting to start and let it finish off any that are already running. Come back 5 hours later and tell each node to shut down before turning the power off and on again. I don't know what caused this node to lock up but suspect the micro SD card is probably worn out. To change them I would have to open up the Edge Cluster find the particular node and replace the SD card and then reinstall things. At the moment its back after the power cycle so I will see if it has any further issues.

You can run Pis using network storage, but I don't, I took the simpler solution of using the micro SD card for booting and local storage.