Most hosting companies advertise resource limits. Very few of them enforce them.
The standard approach is soft limits — a threshold that triggers a warning email, or a control panel flag that a human might act on. These are not guarantees. They're notifications.
We do something different: kernel-enforced hard limits using Linux cgroups.
What a cgroup Is
A cgroup (control group) is a Linux kernel feature that lets you assign processes to a named group and attach hard limits to that group: maximum RAM, CPU shares, I/O bandwidth, number of processes. The kernel enforces these at the scheduler level — not in software, not in a monitoring agent, but in the same code path that allocates resources to every process on the system.
Every customer account on our servers runs inside its own cgroup. Your PHP workers, your database connection, your cron jobs — all of them are children of your cgroup.
What Happens When You Hit a Limit
If your PHP process tries to allocate more memory than your plan allows, the kernel kills it. The request fails. Your error log records the event. Other customers are unaffected.
This sounds punishing, but it's the honest behaviour. The alternative — letting your site consume memory until the server runs out — is what causes the cascading failures you get on oversold servers.
CPU Enforcement
CPU limits work on shares, not hard percentages. Your cgroup gets a guaranteed slice of CPU time. If the server is idle, you can burst above your share. If the server is busy, you get exactly your share and no more.
This means your site doesn't slow down because someone else's site is doing something CPU-intensive. Their CPU usage is capped. Yours is not their problem.
Why Soft Limits Don't Work
A soft limit depends on a monitoring process noticing an exceedance, deciding to act, and successfully terminating or throttling the offending process. That chain has latency. During that latency, resources are being consumed that belong to other customers.
A hard limit has zero latency. The kernel enforces it before the resource is consumed.
The Thing You Can Verify
If you're on our platform, you can run site info from your shell to see your current limits and usage. We don't hide the numbers behind a dashboard that only shows aggregates. Your cgroup limits are your actual limits.
This is what "no overselling" actually means at the technical level. Not a policy. Not a promise. A kernel constraint.