How to Evaluate VPS Providers: 5 Criteria That Predict Real Performance and Total Cost

Provider comparison pages are dominated by marketing spec sheets: vCPU counts, RAM, “unlimited” claims. Those numbers predict very little. What actually predicts whether a cheap VPS performs and what it truly costs comes down to seven measurable things you can test yourself in about an hour — before you pay for a year.

1. CPU Steal Time

Steal time is the percentage of CPU cycles your VM wanted but did not get because a neighbour was using the physical core. On budget nodes, sustained steal above 10% means your 2 vCPU is behaving like roughly 1.5. Run top or vmstat 1 under load and watch the st column. A provider with 2% steady-state steal is worth paying more for than one advertising an extra core.

2. Disk IOPS and Latency

Storage is the most common bottleneck on cheap plans. Test with a simple sequential write and a small random-write loop. Watch for:

• Sequential write below 100 MB/s — weak for a modern SSD plan
• Random 4K latency above 1 ms — poor for databases
• Wide variance between runs — indicates shared, contended storage

Run the test at three different times of day. Consistency matters more than peak numbers.

3. Network Throughput and Peering

A 1 Gbps port claim is meaningless if the provider’s peering is poor. Measure actual throughput to several destinations and check latency to your real audience. A plan with 500 Mbps and good peering beats a 1 Gbps port that routes through three congested exchanges.

4. Real Usable RAM

Check free -m immediately after boot. Some providers’ base OS images consume 300–500 MB before you install anything, and control-panel agents can add more. A “1 GB” plan with 400 MB already gone leaves you 600 MB for your application.

5. Renewal Pricing and Add-On Costs

Cost Item	What to Record
Intro rate	What you pay now
Renewal rate	What you pay after term 1
Backups	Included or $1–$5/month
Extra IPv4	$1–$3 each/month
Bandwidth overage	Per GB rate

Add these up for a 24-month model. The cheapest intro price is rarely the cheapest total. Our provider comparison table shows intro and renewal rates side by side so the second year is never a surprise.

6. Support Response Reality

Open a pre-sales ticket with a specific technical question — for example, how the provider handles CPU fair-use throttling. Measure the response time and whether the answer is specific or canned. Budget providers distinguish themselves here; the ones that answer in ten minutes with real detail are worth a premium of a few dollars.

7. Migration and Exit Costs

Ask two questions: can you take a full disk image, and is there a fee to leave? Providers that charge for outbound transfer or make backups difficult to export are creating lock-in. Test the export path on day one, not when you need to leave.

Scoring the Seven Tests

Test	Pass	Fail
CPU steal	<5%	>15%
Disk random latency	<0.5 ms	>2 ms
Network consistency	Stable across runs	High variance
Usable RAM	>80% of advertised	<60%
2-yr total cost	Within 30% of intro	Doubles
Support answer	Specific, fast	Canned, slow
Exit path	Full image export	Locked in

Any provider that passes five of seven is a reasonable buy. One that fails four is a false economy regardless of price. Start from these seven tests, then compare the current budget providers using the same criteria rather than their marketing copy.

Putting the Tests in Order

Not every test is worth the same weight. If you only have thirty minutes, prioritise in this order: renewal pricing first (it determines total cost), then CPU steal and disk latency (they determine daily performance), then network and support. The remaining tests matter, but they rarely change a purchasing decision on their own.

Sample Scoring Sheet

Criterion	Weight	Provider A	Provider B
2-year total cost	30%	8/10	6/10
CPU steal	20%	9/10	7/10
Disk latency	20%	7/10	8/10
Network consistency	10%	8/10	8/10
Support quality	10%	6/10	9/10
Exit path	10%	9/10	5/10
Weighted total	100%	7.9	7.0

Adjust the weights to match your situation: a bandwidth-heavy project should weight network higher, a database should weight disk latency higher. The point is to make every provider compete on the same terms.

Common Mistakes in Provider Comparisons

• Comparing intro rates across providers. Anchor on renewal rates instead.
• Ignoring the base OS footprint. Two providers selling “1 GB” can leave you 400 MB or 700 MB free.
• Benchmarking once. A single run at a quiet hour flatters every host. Repeat at peak.
• Trusting “unlimited”. Find the fair-use or port-speed limit behind the word.
• Skipping the exit test. Export your full image on day one to confirm you can leave.

What to Do With the Results

A provider that scores well on cost, steal and disk latency is the right default for budget buyers. Support and exit path are tie-breakers. Run the seven tests once per candidate, score them on the same sheet, and the decision usually becomes obvious within an afternoon.