If you've bought shared hosting and your site runs slowly during business hours, you've experienced overselling. It's the most common cause of poor performance on budget shared hosting, and almost no hosting company talks about it openly.
The Maths of Overselling
A typical shared hosting server might have 64GB of RAM. The host sells 1000 customers on plans advertised as "1GB RAM". That's 1000GB of claimed RAM on a 64GB server — a 15:1 oversell ratio.
How can they do this? Because most customers don't use their full allocation at any given moment. A WordPress blog that gets 100 visitors per day uses almost no RAM overnight. The server has plenty of capacity at 3am.
The problem is peak hours. At 2pm on a Tuesday, when half those WordPress sites are serving real traffic simultaneously, the aggregate demand exceeds physical capacity. The server slows down. Requests queue. Sites time out.
Why Hosts Oversell
Economics. If you sold a 64GB server at honest allocations — say 200 customers with 256MB each — you'd need to price plans significantly higher to cover the server cost. By overselling 5:1 or 10:1, you spread the fixed cost across more customers and can price plans at £2/month instead of £10/month.
The bet is that average load will stay well below peak capacity most of the time. Usually it does. But "usually" isn't a guarantee.
How You Detect It
The clearest sign of an oversold server is TTFB (time to first byte) that varies dramatically by time of day. If your WordPress site responds in 200ms at midnight and 3 seconds at noon, the server is under load during business hours.
Test it: use a website monitoring tool to check your TTFB hourly for a week. If there's a clear pattern correlating with business hours or time zones, you're on an oversold server.
The Soft Limit Problem
Many hosts use "soft" resource limits — PHP memory limits set in configuration files — rather than kernel-enforced hard limits. A soft limit can be exceeded when the enforcement mechanism (often a cron job or monitoring agent) hasn't caught up yet. By the time action is taken, the damage is done.
Hard limits, enforced at the kernel level via cgroups, don't have this problem. The kernel refuses the resource allocation before it happens.
What We Do Instead
We don't oversell. We cap servers at a fixed number of slots, publish live capacity on our homepage, and pause signups when a server is full. We also enforce limits at the kernel level, not in configuration files.
The result is higher infrastructure costs per customer. We offset that by keeping our operation lean and our tooling custom-built. The price difference between our plans and an overselling competitor's plans isn't profit — it's the cost of actually delivering what we promise.