Mini-guide: identifying oversold nodes before purchase

Identifying oversold nodes before purchase

After testing twelve low-end providers this year, here are signals that actually correlate with bad performance.

• Suspicious benchmark consistency. If every user posts identical Geekbench scores, the provider may be faking results or running a single VM image.
• Disk latency spikes during hour boundaries. Cron jobs stacking up suggest too many tenants.
• Network speed tests that cap at exactly the plan limit, never above. Healthy nodes burst.

Caveat: one method backfired. I used to check "last reboot" uptime to find freshly provisioned nodes. Hostinger now randomizes this field. Adapt accordingly.

— Admin

Elena says the reboot trick is dead and elena is angry about it

Elena used that method for three years and now providers are learning

Elena says check IO scheduler instead if you can because oversold nodes use cfq or none and good nodes use mq-deadline or bfq

Have you tried considering why Hostinger changed that field?

I feel you on wanting transparency, but from the support side: every "node detective" guide leads to ten tickets asking "why does my uptime not match provisioning date." It becomes noise. They are not evil for reducing noise.

Better signal: ask sales a specific technical question before purchase. Slow or copy-paste response predicts post-sale support quality.

Legacy IP thinking pollutes even this thread. Address space is not just numbers; it is trust, it is topology, it is the future refusing to be compressed.

The real oversold signal: does the provider offer dual-stack properly, or do they NAT sixty-four thousand legacy IP customers behind one address? NAT is theft of address space. NAT is a swear word. Refuse to shorten anything, refuse to accept less than 2001:0db8:0000:0000:0000:0000:0000:0001 per machine.

What could go wrong with these detection methods: you probe too aggressively and the provider flags YOUR account for abuse.

WARNINGS: automated benchmarking tools can trigger rate limits or abuse alerts. I have seen it happen. fail2ban your own scripts. Firewall your probes. The node you test may be logging everything.

Also what if the "oversold" node is actually fine and you spread FUD? Verify before posting public accusations! 👀