# Diagnosing Overhead

**URL:** <https://discuss.ray.io/t/diagnosing-overhead/1282>\
**Category:** RLlib\
**Created:** [March 16, 2021, 5:28pm UTC](https://discuss.ray.io/t/diagnosing-overhead/1282 "2021-03-16T17:28:14Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![jsuarez5341](https://avatars.discourse-cdn.com/v4/letter/j/50afbb/32.png) [@jsuarez5341](https://discuss.ray.io/u/jsuarez5341)\
**Post date:** [March 16, 2021, 5:28pm UTC](https://discuss.ray.io/t/diagnosing-overhead/1282/1 "2021-03-16T17:28:15Z")

</div>

sample\_time and learn\_time below do not sum to wall clock time

```
stats = self.train()
sample_time = timers['sample_time_ms']
learn_time = timers['learn_time_ms']

```

I’m seeing about 10-20% overhead. Does RLlib track and additional timing data anywhere that could help diagnose the source?

---

<div class="post-metadata">

**Author:** ![Dmitri](https://sea2.discourse-cdn.com/flex020/user_avatar/discuss.ray.io/dmitri/32/657_2.png) [@Dmitri](https://discuss.ray.io/u/Dmitri)\
**Post date:** [March 17, 2021, 1:58am UTC](https://discuss.ray.io/t/diagnosing-overhead/1282/2 "2021-03-17T01:58:57Z")

</div>

cc @sven1977 -------