One of the sites that I monitor that was fairly quick previously has seen very intermittent speeds since mid December. We’re using SpeedCurve (which uses Web Page Test).
As you can see in the attached image - from mid December both Firefox and iPhone took a big hit in speed. Our Devs don’t know of any code or server changes that could have caused this.
Here are two examples of slow tests (1 firefox and 1 iphone)
The Firefox test has both fast and slow cases in the same test so it’s sort of possible to see what is happening but it doesn’t look like it is content related. The browser seems to go out to lunch between completing the TLS negotiation and fetching the resource. Hard to say if it is a machine issue or something going on with cert validation but whatever is happening looks to be local to the machine and not server related.
Do you have a “good” test result for the iPhone case? All 3 runs in the test you had looked like they had similar performance with no obvious gaps so I’d need something to compare it to to see what changed.
I’ve noticed that some time in the last month or so Chrome’s started doing OCSP checks on EV certs (or at least they’re showing up in the waterfalls) and I’m seeing this push some waterfalls out.
I’m not sure whether I’ve just seen the increased OCSP checks for Symantec signed certs or for other CAs too.
As the iPhone is really Chrome, could that be the source of the change here?
When you say the Firefox tests seems machine related and not server related do you mean that is likely the test machine that has an issue and not the webserver? I’ve seen similar tests on more than one machine running WebPageTest.
In the iPhone case the change is due to the HTTPS negotiation for the base page now doing OCSP validation checks where it wasn’t before (which is what Andy mentioned also). Here are the two tests next to each other: WebPageTest - Visual Comparison
Did you recently move to an EV Certificate where you weren’t using one before or were you possibly using OCSP Stapling before and are no longer? The current certificate is definitely EV and it looks like the certificate itself was issued close to a year ago so I’d be more inclined to think it was server config issues with stapling: SSL Server Test: www.uswitchforbusiness.com (Powered by Qualys SSL Labs)
Unfortunately I can’t see if the old tests had stapled certificates.
Firefox in theory could be the same or a similar issue because the delay is also right where certificates would be validated but I’d expect to see the certificate checks if that were the case (like we see with iPhone). If you can reproduce the Firefox slowness with tcpdump enabled (either on the public instance or on Speed Curve) then that would prove for sure one way or another.
As far as it being something on the machine, it could just be some decisions the browser is making but usually when there are big gaps with no CPU and no Bandwidth it is a machine issue but certificate validations if they aren’t exposed in the waterfall could also cause the same behavior.