If I wanted to compare a websites own RUM data vs ‘global’ RUM data e.g. page load, is there anywhere that tracks this over time? e.g. median onload for March 2022.
It’s a little like this report, but I’d like to know what the figure is in seconds.
At the moment, I could say a site is in the top 30% of sites out of 1 million since it has a ‘good’ onload score of less than 1 second. But it just seems a little long-winded for non-performance folks to understand. I think “site x has a median load time of 0.4s, this compares to 1.5s for the top 1 million sites on the web” is much easier to understand and explain.
WebPageTest compares the performance of all websites with each other to calculate the speedIndex. This is a great metric but it looks like detailed docs are MIA at the moment:
IIRC the metric is a little more sophisticated than just the median, but Pat Meenan can say more.
However, comparing with “all” the websites is, in my view, somewhat limited. The dataset contains a lot of very similar websites such as YouTube channels or WordPress sites. Of course, this doesn’t make the data invalid, it just means you’re not doing like for like comparisons.
For my own customers, I extended the idea of slices that the original httparchive.org had so that they can see how their website performs within a sector or with selected “benchmark” or best practice sites.
Actually, while I’m at it, it’s worth reiterating what a great resource WebPageTest is for anyone who’s genuinely interested in the performance of their website. It takes the guess work out of the process and provides detailed information for improvements.
Are you specifically after
I attached a similar report for LCP, showing LCP in milliseconds by percentile over time. If it is what you’re after, I can see if I can extend it to show
onload and group by rank (most popular 1000 sites, 10000 sites…etc). Currently it queries the entire data set according to CrUX.
This way you can say that your website at 1500ms is between the 10th and 25th percentile.
APPROX_QUANTILES(p75_lcp, 1000 RESPECT NULLS)[OFFSET(percentile * 10)] AS p75_lcp
UNNEST([10, 25, 50, 75, 90]) AS percentile
date >= '2022-01-01' AND
date <= '2022-03-01' AND
device IN ('desktop', 'phone')
I can share some of the Akamai mPulse RUM data, though please read the caveats below.
Here’s a histogram of all Page Load times across all of our customers for a recent 24 hour period:
- 25th: 1,375ms
- 50th: 2,875ms
- 75th: 5,375ms
- 90th: 10,125ms
- 95th: 15,125ms
Some caveats with this data.
- This is just a representation of Akamai mPulse customers, not the top N websites
- This data is not normalized by website, meaning more popular websites may be over-represented. There are NN thousand domains represented in this data.
- This data shows Page Load times, across both landing/home pages and in-site navigations, across all browsers.
- This does not include mPulse websites that have enabled mPulse Single Page App measurement support (which mPulse measures differently from
onload event timings)
- This data is from a single 24-hour period on Thurs Apr 09 through Friday Apr 08 (ET)
- This data only includes
visibilityState=visible (so not tabs loaded in background) and
navigationType=navigate (so not reloads, back-forward, or BFCache navs)