What Is a Non-Oversold VPS? How to Tell If Yours Is Oversold

RemarkableCloud Team

What is a non-oversold VPS?

A non-oversold VPS is a virtual server whose host never sells the same physical resource twice. If the physical machine has 64 cores and 256 GiB of RAM, a non-oversold host sells at most 64 vCPU and 256 GiB across all the servers on it, so the resources on your plan sheet are physically reserved for you. An oversold host sells 150, 200, or 400 percent of the hardware and bets that customers will not use their allocations at the same time.

That bet works on quiet days and fails exactly when it matters: traffic spikes, backup windows, Black Friday, the moment your post goes viral. Overselling is why two VPS plans with identical specs on paper can perform completely differently in production.

How does VPS overselling actually work?

Overselling works because most servers idle most of the time. A typical website uses a few percent of its CPU allocation on an average day, so a provider can place 30 “4-core” servers on a 64-core machine and nobody notices at 3 a.m. The hypervisor time-slices the physical cores among everyone who wants them.

The problem is statistical, and it is not on your side. When load rises across tenants at once, the hypervisor has to ration. Your “dedicated” 4 cores become a queue ticket. The symptoms are distinctive:

• CPU steal time. Your OS wanted the CPU and the hypervisor gave it to someone else. Anything above 1 to 2 percent sustained is a rationing signal.
• Inconsistent benchmarks. The same workload takes 40 seconds at dawn and 90 seconds at peak, on the same server, with nothing changed.
• Fast disk, slow app. NVMe numbers look great in a synthetic test, but real requests stall because CPU and I/O are contended by neighbors.

How can I test whether my current VPS is oversold?

Check steal time first, because it is the one metric a host cannot hide. Run top or vmstat 1 on a Linux VPS and look at the st column. On a non-oversold server it reads 0.0 essentially always. If you see sustained steal during your busy hours, your cores are being shared beyond capacity.

Second, benchmark at different hours. Run the same short CPU benchmark (for example sysbench cpu run) at 4 a.m. and at your traffic peak. On honest hardware the results match within a few percent. A double-digit gap means you are competing for the silicon you paid for.

Third, read the terms. Oversold hosts rarely say “oversold”; they say “fair use”, “burstable”, or “up to X cores”. Reserved resources need no hedging language.

Why do oversold plans look like such good deals?

Because overselling is the discount. Selling one core three times cuts the hardware cost per customer to a third, which is how the market floor got to a few dollars a month. The price is real; the resources are conditional. You are not buying 4 cores, you are buying a lottery ticket for up to 4 cores whose odds worsen every time the host onboards a neighbor.

There is nothing wrong with that trade for a hobby project. It is the wrong trade for eCommerce, client sites, or anything where a slow checkout costs real money, because the plan fails precisely during the high-traffic moments that pay for everything else.

What does non-oversold hosting look like in practice?

Every RemarkableCloud plan runs on enterprise dedicated infrastructure in US data centers with resources that are never sold twice: what the plan sheet says is what the hypervisor reserves. Shared CPU plans share physical cores only up to their real capacity, and Dedicated CPU plans pin cores exclusively to your server. Both families run 100% NVMe at +3.00 GHz.

The proof is operational, not promotional: steal time on our fleet reads zero, benchmarks repeat within noise at any hour, and we publish a 500% SLA that credits 5 hours back for every hour down, from the first minute. A host that oversells cannot afford that SLA, because contention-driven slowdowns would bankrupt it.

If you want to see the difference on your own workload, managed VPS plans include free migration, so the test costs you nothing: we move the site, you run your benchmarks at peak, and the 30-day money-back guarantee covers the exit if we are wrong.

FAQ

Does “dedicated resources” always mean non-oversold?

No. “Dedicated” in marketing copy often refers only to RAM, while CPU remains shared beyond capacity. The reliable test is behavioral: zero steal time and repeatable benchmarks at peak hours. Ask the host directly whether CPU is sold beyond physical capacity and keep the answer in writing.

Is a shared-CPU plan the same as an oversold plan?

No. Shared CPU means multiple servers legitimately share physical cores within the machine’s real capacity, which keeps prices reasonable while performance stays consistent. Overselling means the host sold more capacity than the machine has. A shared-CPU plan on a non-oversold host outperforms a “dedicated” plan on an oversold one at peak.

How much steal time is acceptable?

Essentially none. Brief spikes under 1 percent during live migrations are normal in any virtualized environment. Sustained steal of 2 percent or more during your business hours means your host has committed your cores to someone else, and no application-level tuning will fix it.