“VDS” and “VPS” are used interchangeably by most providers, and by most buyers, but the words describe different isolation guarantees. A cheap VDS (virtual dedicated server) usually means fully reserved or dedicated resources with light-to-no oversubscription. A cheap VPS usually means a share of a larger node with overcommitted RAM and CPU. The price gap between the two is small on paper and large in practice, and it shows up in five specific performance behaviours that decide whether the plan works for your workload.
CPU steal time: the metric that explains the price
Steal time is the percentage of time your virtual CPU was ready to run but the hypervisor scheduled another tenant instead. On a well-run VPS node it stays low; on an oversubscribed budget node it can sit at 10% to 30% under load, which means your application runs at 70% of the speed you paid for. A VDS with dedicated cores should show steal near zero, because no other tenant shares the physical core. Reading steal is simple: run top and look at the st column, or read /proc/stat for the steal counter.
The commercial consequence is direct. If a VPS at $7 has 20% steal under your peak load, you are effectively paying $8.75 for the performance you receive, which may still be a good deal, but it is no longer cheaper than a $9 VDS by as much as the listing suggested. Measure steal at your peak hour, not at 3 a.m. in the host’s timezone, and measure it for a week rather than a minute. A single low reading proves nothing.
Memory ballooning and swap behaviour
Overcommitted RAM is handled by ballooning, where the hypervisor reclaims memory from guests that are not using it. That works fine until several tenants need their memory at the same time, at which point guests get squeezed into swap and latency rises sharply. On a VPS you may see swap in use and page-in rates spike with no obvious cause in your application. On a VDS with reserved memory, that pressure should not occur from a neighbour’s behaviour.
Watch vmstat 1 for the si and so columns during your busiest period, and check how much swap the host provisions relative to RAM. A plan with 2GB RAM and no swap leaves you no headroom for a soft memory spike. A plan with reserved memory and 2GB swap gives you a cushion. Neither is inherently better, but the cheap plan’s failure mode is a hard OOM kill, and that is a production incident rather than a slow page.
Disk IOPS: the hidden ceiling on cheap plans
Storage is where budget providers economise most aggressively. Many cheap VPS plans run on shared NVMe or SATA arrays with a per-tenant IOPS cap, and the cap is often not published. You can have a 4-core, 8GB plan that runs a database at 300 IOPS while a 2-core VDS on dedicated NVMe delivers several thousand. For a static site this is irrelevant; for a database-backed application or a build server, it is the difference between a plan that works and one that does not.
Benchmark before you commit, using fio for random 4K reads and writes and a simple dd test for sequential throughput, and run the random test three times to see variance. Then look at the ratio of IOPS to price. A cheap plan with high sustained IOPS and low CPU steal is a better buy than a cheaper plan with higher core counts and a throttled disk, because database workloads are usually IO-bound before they are CPU-bound.
Network throughput and noisy-neighbour effects
Network is shared at the host level even when the CPU is not. On a budget VPS, the physical NIC and the upstream link are shared across many tenants, and a single neighbour running a large transfer can degrade your throughput and, more importantly, your latency and packet loss. A VDS with a guaranteed slice of bandwidth, or a dedicated port, isolates you from that. Whether it matters depends on whether you serve HTTP or run latency-sensitive traffic such as game servers, VoIP, or real-time APIs.
Test with a sustained transfer to a nearby and a distant endpoint at peak and off-peak times, and record latency percentiles rather than averages. Averages hide the tail latency that users actually feel. If your p95 latency doubles at peak, the plan is sharing more than the ad implies, and the fix is either a better plan or a CDN in front, which costs money and adds moving parts.
Where you land depends on workload shape, and it is worth looking at the budget plans we currently compare with reserved vs shared resources listed side by side, because the difference between a shared 100 Mbps slice and a guaranteed one rarely appears in the plan name.
The control plane: reboots, snapshots, and console access
The performance differences above are the ones that show up in benchmarks. The ones that show up in operations are subtler: how long a reboot takes, whether snapshots are instant or queued behind a job, and whether you get out-of-band console access when the guest will not boot. Budget VPS plans frequently queue snapshots and offer no console, so a failed boot means a support ticket and a wait. A VDS, even a cheap one, usually includes a console and prioritised storage operations.
Cost this in incident time, not in dollars. If a bad kernel update leaves you unable to boot and your only recovery path is a support ticket with a four-hour response, the difference in recovery time between the two plans dwarfs the monthly price difference. For anything you would call production, console access and snapshot rollback are not luxuries.
When a cheap VPS is the better buy anyway
None of this means a VDS is always better. For a static site behind a CDN, a small cron-driven job, an internal tool used by a handful of people, or a staging environment, the noisy-neighbour risks are acceptable and the price difference is real money. The rule of thumb: if your workload can tolerate a slow page and a reboot, a cheap VPS is fine; if it cannot, pay for reserved resources and treat the premium as insurance.
Consider the shape of the workload too. Bursty workloads that idle most of the day benefit from shared CPU, because you are buying peak capacity for a fraction of the cost. Sustained workloads with steady utilisation are the opposite: you are paying for the average, and a dedicated plan priced per core is often cheaper per unit of delivered performance. Match the billing model to the shape of the load, not to the headline number of cores.
A practical decision table
Score your workload on four questions: does it need predictable CPU under load, does it need low-latency disk, does it need stable network latency, and does a failed boot cost you money? Two or more yes answers points to a VDS or a VPS plan with reserved resources. Zero or one points to the cheapest shared plan you can find, ideally with a short billing term so you can move if performance disappoints.
Whatever you choose, measure it in the first week and keep the numbers. Steal time, IOPS, and p95 latency give you a defensible answer when the next renewal quote arrives. Review our current VPS and VDS listings and compare plans on measured performance rather than on the core count in the plan name.


