dru 🦀

kubernetes on riscv showing all pods in all namespaces and all nodes

grafana dashboard; currently tweaking my custom nvme exporter

microcluster is now fully functional.
Kubernetes across all nodes, local container registry, Grafana + Prometheus, custom exporters & custom certs with everything built & stored locally because upstream support is iffy at best & tends to vanish without warning.

all entirely on RISC-V

finally got grafana running on risc-v

homelab is coming along nicely 👌🏼

6 days since the last riscv microcluster update and things are going fairly well. ive been battling grafana on riscv today. its currently winning.

however kubernetes on riscv is full functional, as is docker, lxc and a few other cool bits

learning loads which is always a bonus.

took my own advice tho; RTFM! downloaded the datasheet & discovered these boards ~will~ boot from nvme and the bootloader can be flashed to spi; which means i only need 1 sd card, once.

plus, i fixed my corne keyboard!

ups and downs my friends :)

my homelab :)

alas, one of the nodes was faulty. sd card slot would not read any of my seven cards. tried shimming it, holding it in, reseating a million times, different angles, nothing. these boards are useless without the first boot with sd; otherwise no os install

the last rv2 node was delivered this afternoon a day early, so my risc-v microcluster is now 90% complete. ordering an 8gb r2s (4x lan ports) on friday which will serve as a backplane across these nodes & will also run containers/microservices; which explains why i want 8gb ram on a router board :)

i intend on finalising the build with an orangepi r2s, which boasts four lan ports. i will connect lan1 of each new node (lan0 is reserved for direct uplink to switch) to the r2s to create a backplane for either containers or microservices & it will leave 1 lan port free for a ch32v microcontroller

adding nvme

added rtc battery

old vs new. wyse 3040 x86 2gb ram vs orangepi 8gb ram riscv

upgrade 80% complete

bit of a serious upgrade too as im gaining 8gb ram (4-6gb increase), 2x nvme and 2x gigabit lan ports across each node.

end of an era, breaking down my microcluster to upgrade to riscv

nvme & luckfox pico max

just waiting on delivery now. upgrading the entire microcluster to risc-v will satisfy two of the critical points in my roadmap; ill be doing rust anyways but next month ill be doing it all on risc-v.

should be fun too, another rabbit hole to venture down :)
xo

cm5 io board with a sizeable nvme cooler

its the first cluster i built and while it was great for learning its time to upgrade. ive been thinking about it since november and i recently stumbled on some risc-v sbcs that have 8gb ram, dual lan AND dual nvme so that was me sold. i grabbed three of them to replace each node

microcluster, cm5 4gb ram 128gb native nvme & 2 wyse 3040s each with 128gb external nvme via usb. luckfox pico max 256mb sits between them

2026. i have a roadmap i want to achieve this year and the two biggest points on it are further my rust knowledge and delve into risc-v. my lovely wife bought me ultimate rust for systems programming for christmas so i have that to work through & i have decided to upgrade my microcluster to risc-v

in lieu of turning 36 today

happy new year to all my followers and friends 🥳 xo

finally got a micro llm running on the luckfox max! this board only has 256mb of ram so it was no easy feat trying to find a model that would work, gpt-mini q2 & q6 work, but 3/4/5 dont... odd.

it outputs absolute garbage; but proof of concept is there.

3.4t/s & 1.1t/s is pretty impressive also.

epic gift from my beautiful wife, ive wanted this rust tome for a very long time.

our rescue rottweiler, Jess

the best christmas present ever 🫶🏼

grafana dashboard showing data from my media center, microcluster and ai node

sat here tonight and messed around with grafana because ive never really used it. got a nice dashbord now pulling data from my cm4, microcluster and one of my ai nodes. the cooler i installed is definitely working on the ai node as they usually run at about 50c+.

luckfox max mounted inside a teucer nvme cooler

luckfox max inside a teucer cooler, it doesnt look too bad. id rather have not used zipties but we cant have it all

in the end it was indeed, a bodge job with zipties. couldnt get it to mount without over stressing the board & needed the soc to touch the underside of the top section of this cooler, used copper pads for that.

it doesn't look too bad. need them to lower temps as these boards run warm. func > form

luckfox max sbc, various tools and a teucer nvme cooler.

i had an idea of mounting a headerless luckfox pico max inside an nvme cooler. i have a cooler spare and another luckfox dropped today, i need a few of them anyways for another project but just want to see if i can somehow mount it inside. it might have to be a bodge job with zipties tho.

wroom32, supermini c3's, rp2040plus and esp s3

esp s3

rp2040plus

luckfox pico max & 5v poe splitter

microcontroller drawer.

recently got a book on esp32 & risc-v, so that will be interesting to work through with the supermini c3s... definitely going to make use of the s3s, edge nodes for another project

luckfox pico max runs well off a 5v poe splitter too, so that solidifies another idea ive got

3 monitors with grafana that visualises microcluster data, btop on the left and right monitor also showing data from all servers in my network

always a joy to sit here.

finally managed to add my microcluster to the main network without "no route to host". issue was stale arp tables in my office switch, i have a few daisychained together & had a hunch this was the reason. finally added the ip's as reserved addresses in my router & can now view grafana from my phone

btop showing 1c cpu temp

am1i legacy rig athlon 5350 8gb ram nvidia k2000d 2x 500gb hdd win xp x64 & debian bookworm nano itx above the am1i intel n3350 4gb ram 64gb emmc freebsd 12 x64

went back to debian + xfce on my legacy rig, kde neon is lovely but very heavy. legacy rig usually runs xp x64 for metin2 dev (+ playing fallout 3 / skyrim).

1c cpu :D

psu is 160w pico, but never draws > 100w, with current electric prices gotta keep that wattage down. entire homelab uses ~240w.

luckfox max module is here, ive got a few of these coming from different sources.

rockchip rv1106 soc
- cortex a7 @ 1.2ghz
- built in npu with 1 tops compute
256mb ddr3l ram
256mb nand flash
10/100 ethernet
isp support
os support - ubuntu/buildroot/alpine

going to have lots of fun with these :)

im happy with it so far tho. originally it was just a microcluster for k8s, rust & microservices but i decided to put mcus on it and use it for mcu development too. just needs polishing, e.g; cable ties arent ideal but work for now.

ill write an indepth blog post when i get the time, promise :) x

the handles are from old rack handles i took off my liebert gxt3000 ups, as i dont need them. its a bit of a mess right now as i removed the pi & im still waiting on an 128gb nvme to finish the top section, and squeezing a luckfox pico max inbetween them for light AI tasks and more mcu development.

originally it had a rk118tv box that i patched armbian into, but replaced it with a cm5 as it has more ram. everything is poe*, i managed to squeeze poe splitters behind the wyse nodes as they dont support native poe.

* except the cm5, just waiting on a hat.

another project im working on is my microcluster thats comprised of a cm5 8gb + 128gb nvme, and 2x wyse 3040s running debian. the wyse nodes have usb nvme drives attached to them, as well as a microcontroller on each wyse for mcu development, all housed in a modified 3 bay 3.5" hdd caddy