guys pls help home lab on a single host. allocaetd vCPUs across VMs add up to well over the physical core count, and memory is mostly committed. things feel slow at times, but the host dashboard doesnt look maxed out.
how do ppl approach sizing and overcommit for small setups? its a single 8-core box with 64 GB of RAM... about ten VMs, most with 4 vCPUs. any ideas??
oof to add some contxet: memory is where Id be stricter. CPU time can be shared, but if the host runs out of RAM it has to balloon, compress or swap, and everything slows down a lot 😂 Keep teh total allocated RAM below physical memory minus what teh host needs, unless you know ur hypervisors memory sharing works well for your workload.
@late_smith cut most Linux VMs to 1-2 vCPUs, gave teh firewall a dedicated amount, and freed some RAM. Steal time dropped to near zero and teh Windows VM feels normal again. good luck man!
CPU overcommit is normal and usually fine, but oversized VMs hurt. A VM with 4 vCPUs needs the scheduler to find time for those vCPUs, and on a busy host that causes waiting (CPU ready time or steal) even when average usage is low. Start most Linux servers at 1-2 vCPUs and increase only if they actually hit their limit.
oof hey, @sami.halvorsen Id disagree a little with starting everything at 1 vCPU. Some workloads, like a Windows desktop VM, get noticealby laggy with a single vCPU even when idle, because of background tasks. 2 vCPUs is a safer floor for anything interactive; 1 is fine for small Linux services. good luck man!
Sign in or create an account to reply.
