Sandcastle profiling category?

[mramato]

There's a Profiling category in Sandcastle with a single example about profiling half-dome (and it has no associated screenshot). Was this merged into master by accident? It seems too developer specific to be included in Sandcastle proper as part of a release.

[pjcozzi]

From #2102

I'd still rather not have these here; it is yet another place we need to remember to update. Perhaps just separate Sandcastle examples would be OK - long-term, we may have a label for examples not included in the final release but used for testing and profiling. Or perhaps there is a way another way to get a time interval from the network profile? Or just not submit anything for now.

[mramato]

It should be easy to put any dev only examples in a "gallery/debug" directory and then not include them in the release zip. I have a few gists I currently use that I would probably throw in there if we do this. I'll look into it.

[shunter]

I would say just get rid of the profiling example, and move it to a gist as Matt does (I have done the same as well). I don't see the benefit in maintaining one-off test cases like these in the repository directly. It's not a useful example or demo.

[pjcozzi]

Removing is OK with me.

[kring]

I'm the one that merged that, sorry.

[mramato]

No big deal. I was under the assumption we really wanted it in there; sounds like it's no big deal to just remove it then.

[chris-cooper]

Can I suggest some consideration is given to profiling a standard set of scenarios like the one removed? Ideally this would be part of the testing framework to help identify performance regressions. The reason I created the sandcastle example was to be a simple manual first step in that direction. Sandcastle has the advantage that it's visual so you can get a perceptive feel for how long it's taking. I for one was pretty shocked at how long this scene was taking to resolve on a laptop (see #2102)

[pjcozzi]

We do some manual performance testing each release with CPU- and GPU-bound Sandcastle scenarios to make sure that performance (1) doesn't degrade, or (2) degrades for a reason.