First off I’d like to say that the new video capture option is an outstanding feature, however, I am curious if you or anyone else have tested (or noticed) the influence it has on the end results.
I have been running tests this evening and noticed that the TTFBs (and thus, the rest of the tests) seem to be 1.5x - 2x slower with the Capture Video option selected. In some cases it is making almost a 1s difference to total load time. From what I can tell, it may be spiking the CPU on the test machine(s)?
Here are some tests that I ran, one of which is a site in development and another site that is on a separate server and is live in production:
Thanks. I haven’t seen that large of an impact before but it could just be because most of the sites I see are a lot slower and the impact is magnified on a fast page.
I’ve been planning to do some tweaks to further reduce the impact of the video capture (it should already be very light but as you saw it’s not non-existent). I’ll see what I can do in the short term that would give the most gain. Looking at the first byte times it looks like the bulk of the increase is in the DNS lookup (i.e. right at the start) which makes me think the initialization/startup of the capture may be happening inside of the measurement instead of outside of it and I might be able to get rid of most of that impact. I also have some pretty good ideas on reducing the overhead for the actual frame grabs which could help.
I’ll let you know when I think I have it improved.
I made some fairly significant gains in reducing the overhead. It’s not completely eliminated but it is almost in the noise at this point.
Your Dev page: Without Video - 1.26 sec fastest run With Video - 1.34 sec fastest run
(the averages actually have the with video faster because of the swing in times)
I have some more radical ideas on making it even lighter weight but those changes are pretty extreme and I’m probably going to hold off on them if I can avoid it.
Let me know if you are still seeing issues or if you see any more problems pop up.
Wow! Pat, never expected a response time like that. Thanks, that’s absolutely awesome. You’re right, that’s definitely in the area where it’s more like a fluctuation than a substantial impact.
Knowing to compensate for a slight bit of wiggle room in at least the TTFB area when I’m tracking my numbers is more than easy enough to do and definitely makes the video even more worth it.