WebPagetest Forums
Page Load discrepency - Printable Version

+- WebPagetest Forums (https://www.webpagetest.org/forums)
+-- Forum: Web Performance (/forumdisplay.php?fid=3)
+--- Forum: Discuss Test Results (/forumdisplay.php?fid=4)
+--- Thread: Page Load discrepency (/showthread.php?tid=15013)

Page Load discrepency - mark.kellett - 08-17-2017 08:46 PM


I have been testing with WPT and am seeing slower responses when compared with other test solutions.

I am using sitespeed.io on a AWS to test against a clients website and I can verify those results with my own tests using chrome.

However the WPT results shows a significantly longer page load even with using same browser, updated bandwidth and running it from London, Manchester and Dulles.

I am running with repeat tests so as to gain an average with cached responses but the time is still out by on average 50%.

I love the features provided by WPT particularly with the Android phones and the RUM testing. So would love to tweak to I can see a similar set of results.

has anyone else seen this?

RE: Page Load discrepency - andydavies - 08-22-2017 07:01 AM

We see similar results from our own product, Performance Analyser (PA) and the London (AWS hosted) WPT instance. (PA is hosted on dedicated VMs in Telehouse, London, rather than AWS cloud)

I've also got a Mac Mini sitting on a VDSL line that produces pretty consistent results (better than Manchester too)

How different are the results? (You can DM me links to the test if you'd prefer)

If the site is UK hosted, then any Dulles test is going to involve additional latency which will affect the results

RE: Page Load discrepency - mark.kellett - 08-22-2017 07:26 PM

The site is hosted in Ireland, but I've seen similar results for Dulles vs London, Manchester. It doesn't seem to deviate too much which is strange considering I would at least expect the latency of Dulles to impact the load time.

RE: Page Load discrepency - datadiggers - 08-23-2017 01:55 AM

This might be due to the number of cores WPT test machines have. Test from Japan and Netherlands (godaddy server). If I remember correctly one more location has beefier server. If results from those locations are more in line with what you're seeing this might be the issue with # of cores.

If I'm not mistaken test agents have 1 core available per test. These days CPUs have 2+ cores so the results with two cores should be more in line with real experiance. SSL, image decoding and parsing may take a lot of CPU time