Virtual servers

Virtualisation means: a software divides a physical computer into several isolated computers. The physical computer is the host. Every isolated computer on it is a guest.

Your VPS is such a guest. It has its own operating system, its own users, its own firewall and its own IP addresses. You see nothing of the neighbour on the same host — and they see nothing of you.

What is listed as a vCPU in the plans is a virtual processor core: a computing thread of the host that its scheduler assigns to your guest. On all three series — NX, NP and NA — there are more vCPUs on a host than it has physical threads. This is called overbooking, and it works because hardly any instance computes around the clock.

As long as the neighbours compute little, you get more computing time than the plan guarantees. Only the base share per vCPU is guaranteed: 25% on NX, 40% on NP. It only takes effect when your burst credit is used up — the computing time your instance saves up during quiet phases. So it does not take effect at all during load peaks lasting seconds, but immediately during hours of full load.

You can read this from a number that every Linux outputs: the steal time in top. It states the proportion of time in which your guest wanted to compute and the host made it wait. If it is permanently more than 10%, the wrong series was chosen — not the plan too small. We do not state our own average here: it would not be verifiable without a measurement period and the number of observed instances. You can read this number yourself in your own instance within seconds.