I’m going to need to deploy some hardware the 2nd half of 2026. Build-out is going to be across 2 colo sites.
Site 1 will need a minimum of 128 cores (not threads) and 1TB of RAM worth of compute split 25/75 between production and dev workloads.
Site 2 will need a minimum of 64 cores (not threads) and 768GB of RAM split 33/66 prod/dev.
Site 2 is intended to be failover from site 1 in the event of an outage.
Each site will have it’s own HA shared storage solution (DRBD + Pacemaker + Corosync + RoCE)
Storage servers will almost certainly be used/refurb enterprise gear using ECC RAM and new flash / spinning rust.
For the compute tier I’m considering:
- Commodity whitebox servers running AMD AM4/AM5 CPUs with DDR RAM
- Threadripper whitebox servers running ECC
- Used/referb enterprise gear
I’m one of the seven people on the planet with oVirt experience so the plan is to use KVM/oVirt for virtualization/management.
The environments will support pretty standard production services like DNS, mail servers, web servers, Rocket.chat, GitLab etc along with dev lab workloads including VPN, nextcloud, kubernetes, load testing etc.
The stack is meant to support moving a startup from the boot-strap phase to seed stage where work will be engineering heavy.
Curious what peoples thoughts are on compute hardware selection.
Working at pre-IPO Racskpace back when their focus was managed hosting plus my own experience over the years suggests that commodity whitebox servers with commodity DDR RAM is 99.99% safe. I also know that arguing that for “production” workloads is nearly the same holy war territory as no-beans vs beans chili.
Thanks in advance for any feedback/suggestions!