Homeprod Upgrades: waiting…

This weekend was supposed to be the weekend that a significant amount of work to our “homeprod” (our home datacenter, that’s too load-bearing to be considered a homelab), but alas, that did not happen.

What’s on the agenda? Well, fully 20% of our house’s night-time power consumption can be attributed to one machine: the disk server, which has two Xeon 5650 CPUs in it, some DDR3, and a space heater for a chipset. This machine idles at around 125W, including disks, possibly more. The problem is that since the heat load of it is manageable, and for most of the year due to our solar PV array it’s only the night-time consumption that matters, the ROI timeframe of just about anything to replace it is incredibly long.

But a few weeks ago I found what I thought was a pretty great deal on a Xeon 1230v5 board with 32GB of DDR4 ECC memory for about what the memory alone would cost. It’s a workstation board, but I don’t really use the IPMI/iDRAC/iLO features anyway. I foolishly didn’t notice that the 1230v5 does not have an iGPU in it, but for about $50AUD I can buy one that has one and is a tiny bit faster anyway (not that this needs the extra horsepower). This should shave around about 50W off from my testing.

Unfortunately, the cooler I ended up using - my leftover Cryorig m9i - is about 10mm too tall to fit in the 3U case, so I must postpone this until I buy a low-profile cooler… probably from Noctua.

Other things we had planned: I’m going to tear down the Kubernetes cluster, and renumber the subnet it’s on. Why? Because for historical reasons, way back when we first started collecting Unifi hardware, we ran the Unifi controller on LXD on the server. We then had some of the containers (Samba, etc) connect to the home VLAN, but the physical server straddled them. This never got addressed, and the entire cluster still operates in the default VLAN, for no good reason… meaning my work machine is protected from a hacked Windows machine, but not a hacked Linux server (the efficacy of such protection is up in the air - it’s predicated on someone having the ability to compromise a container but not my router, but anyway).

While I’m doing that though, I’m going to rebuild each of the compute nodes to have a secondary SATA SSD in them. I don’t appear to have blogged about it, but for Christmas we upgraded the boot SSDs in all three of our desktops to NVME drives big enough to fit multiple games, so the 500GB SATA drives from mine and Duncan’s desktop and the 250GB one Sabriena was rocking (!!) are now surplus. So I can throw these in the nodes, and then I’m hoping I can set it up so that the disk server will act as the third/backup copy and then things like Plex don’t have to run over EXT4FS over iSCSI, which has caused me no small amount of drama.

Finally, while everything is offline and apart, it needs a good dusting. Desperately. If I have any energy left when all is said and done, I may (emphasis on “may”) do something about the horrific cable managment as well.

Horsham, VIC, Australia fwaggle



Filed under:


Horsham, VIC, Australia

Navigation: Older Entry Newer Entry