Docker compute node build

[HK416]

I need a dedicated docker box. After some research I have come to the conclusion that micro-services using docker and/or Kubernetes would best suit my current and future workloads. To this end, I’m trying to use what I already have and buy as little as possible to reach my goal. At the same time I don’t want to create any future bottlenecks by skimping now if i can help it.

All my data is on a NAS (36TB Freenas) and would be accessed and processed in place by containers running on this new build. I want to be able to concurrently spin up and run as many containers as i need to get the job done. The two boxes would have a dedicated 10Gb link between em. I’m torn between two sets of parts and would greatly appreciate any help or insight to pro and con it out:

• -Corsair RM750X V2 750W (Dead cheap and being sold for a song in our area)
• -Samsung DDR4 2400MHz ECC Reg 16GB x2
• -Supermicro X10SRM-TF (IPMI is the main draw here)
• -Intel Xeon E5-2630 v3 CPU 8 core 16 thread 2,4MHz, 3,2MHz turbo (Found one very cheap)

OR

• -Corsair RM750X V2 750W
• -Asus Prime X370-Pro (Bang for the buck)
• -Samsung 970 EVO Plus Series MZ-V7S250BW 250GB
• -G.Skill Flare X Black DDR4 2400MHz 2x16GB
• -AMD Ryzen 7 1700X 3,4GHz (Core count to price is great)

Cheers

[BGL]

Will this be for labs & learning or production workloads that earn you money (and possibly pay all your bills)?

If for labs, it probably doesn’t matter a huge amount, if for production workloads that run 24/7 things like ECC are pretty much a de-facto requirement. Other than that what’s the max RAM capacity for each board? Labs workloads are generally light on CPU usage but RAM can become a constraining factor when running lots of virtualised workloads.

If you are stopping at a single server then docker onto a bare metal OS install is all you need. If you want to properly master K8’s then you’ll likely want to install VM’s on the single server to create a K8’s cluster so you learn how K8’s manages virtual networks, contianers and pods etc.

[HK416]

Thanks for your reply BGL.

Will this be for labs & learning or production workloads that earn you money (and possibly pay all your bills)?

This will be for production workloads

If for labs, it probably doesn’t matter a huge amount, if for production workloads that run 24/7 things like ECC are pretty much a de-facto requirement. Other than that what’s the max RAM capacity for each board? Labs workloads are generally light on CPU usage but RAM can become a constraining factor when running lots of virtualised workloads.

I take your point. You reminded me of a few things I just glossed over coz of pricing and availability. The Supermicro RAM capacity is 512GB while The Asus is 64GB.

If you are stopping at a single server then docker onto a bare metal OS install is all you need. If you want to properly master K8’s then you’ll likely want to install VM’s on the single server to create a K8’s cluster so you learn how K8’s manages virtual networks, contianers and pods etc.

My thinking was to start with one node and docker then branch or build out as I become reasonably competent with K8s or as my use case and needs change and evolve. My lab is a modestly specked proxmox box and 10 Raspberry Pi model 3 B+ nodes

[BGL]

Cool, let the production requirements guide your decisions. It’s always hard when working to a budget though. Given your options I’d probably opt for EEC RAM and greater expandability over a parts designed for home/gaming use.