Comparisons
Two tools, two questions. Neither column wins — what decides is which question you actually have in front of you.
Tracing vs Profiling
Tracing finds which component holds the time. Profiling finds which code burns the resource inside one component. Reaching for the wrong one is why investigations stall.
Distributed latency: which service, which dependency, which query
What a process is doing during a long span with no children
Instrumentation plus storage; sampling decisions have real consequences
"Where did this request spend its 2.4 seconds?"
CPU and allocation hot spots inside a single process
Anything that happens on another machine, or any time spent waiting
A few percent overhead for sampling profilers; higher for allocation tracking
"Which function is 55% of our CPU?"
| Dimension | Tracing | Profiling |
|---|---|---|
| Unit | A request across services | A process over a time window |
| Best at | Waiting | Working |
| Shows I/O wait | Yes, as span duration | Only in wall-clock or off-CPU profiles |
| Shows a hot function | No | Yes |