Performance vs Scalability
Hop's users complain that it is "slow". That word hides two very different illnesses, and the cure for one does nothing for the other. Before fixing anything, find out which one you have.
The rule
Slow for one user is performance; slow only for the crowd is scalability.
- A performance problem shows up with a single user. The request itself does too much work: a missing index, a wasteful loop, a slow external call.
- A scalability problem shows up only under load. Each request is fine alone, but the system cannot keep up when many arrive together.
A system is scalable when adding resources adds capacity in proportion. Double Hop's servers and it should handle roughly double the clicks. If doubling the servers barely helps, something shared — often the database — is the real limit.
Test yourself with Hop:
| Symptom | Diagnosis |
|---|---|
| Redirect takes 3 s at 4 a.m., no one else online | Performance |
| 20 ms at night, 4 s during the evening peak | Scalability |
| Twice the servers, same peak slowness | Scalability limit in something shared |
The shortcut
Ask one question: "Is it slow when it is quiet?" Yes → performance: look inside the request. No → scalability: look at capacity and shared parts.
Test it quiet
At 3 a.m. a redirect takes 20 ms. Each request is healthy, so this is not performance.
The common mistake is buying bigger machines for a performance bug. A lookup that scans ten million rows is slow on any hardware; an index fixes it for free.
Drills
Answer fast — trust the rule.
A page loads slowly even when only one tester is using the system. What kind of problem is it?
Hop answers in 15 ms with 10 users but in 3 seconds with 10,000. What kind of problem is it?
What does it mean for a system to be scalable?
Hop doubles its servers but peak-time speed does not improve. What is the likely cause?
Remember
- Slow for one user is performance; slow only for the crowd is scalability.
- Shortcut: "is it slow when it is quiet?" Yes → look inside the request; no → look at capacity.
- Trap: bigger hardware for a performance bug, or more servers when a shared part is the limit.