Performance Benchmarks for WordPress 6.7

[mukeshpanchal27]

As we enter the beta phase for WordPress 6.7, I am opening this issue to gather and track all performance benchmarks. This will help us collaborate more effectively, identify potential regressions, and ensure all performance-related data is centralized in one place.

Feel free to contribute with your benchmarks, results, or any findings related to performance in WordPress 6.7.

[mukeshpanchal27]

WP 6.7 Beta 1 Performance Benchmark Report:

I ran benchmarks comparing WordPress 6.7 Beta 1 and the latest released 6.6 branch (6.6.2). I took measurements with the twentytwentyone (TT1) theme using en_US locale, both with object caching disabled and enabled, then again with twentytwentyfour (TT4), This version of WordPress shows a regression in server rendering time from the previous release for TT4.

With object caching disabled, Classic themes (TT1) show a ~3% regression (0.35ms slower) in wp-before-template measurements; LCP remains relatively stable. Block themes (TT4) show a ~6% regression in TTFB (6ms slower) and a ~8% regression (8 ms slower) for wp-total measurements. LCP shows a ~4.5% regression (7ms). Running the tests with object caching enabled shows similar regressions.

Here are the results of the test runs:

Caching Disabled:
Classic theme - https://github.com/mukeshpanchal27/compare-wp-performance/actions/runs/11122906286
Block theme - https://github.com/mukeshpanchal27/compare-wp-performance/actions/runs/11123131063

Caching Enabled:
Classic theme - https://github.com/mukeshpanchal27/compare-wp-performance/actions/runs/11123295178
Block theme - https://github.com/mukeshpanchal27/compare-wp-performance/actions/runs/11123530293

cc. @felixarntz @joemcgill @adamsilverstein @westonruter @swissspidy

[joemcgill]

Thanks @mukeshpanchal27. Repeating what I said in our weekly performance team chat earlier today:

the performance metrics for classic themes looks pretty steady, which is great! For TT4, it seems like we need to investigate the cause of the additional performance regression during template rendering, because adding 10% is not great.

I started trying to investigate the likely cause by looking at the WordPress Code Vitals dashboard and realized that the last commit recorded to that dashboard seems to be this one from five months ago 😱. It's likely that the following commit, r58165, likely broke these.

Fortunately, the Gutenberg tests are still being recorded, and it looks like this commit is worth investigating further. It was synced to WordPress in r58936. Here's the performance run from that commit in WP-dev.

[joemcgill]

I've opened https://core.trac.wordpress.org/ticket/62153 to fix the dashboard issue.

[mukeshpanchal27]

I did the quick compare through xhprof. Here is some finding.

• Use of strpos is significantly more then the previous version.

The compression between 6.6.2 and 6.7 beta1

6.7 beta1 strpos function details:

6.6.2 strpos function details:

[mukeshpanchal27]

Fortunately, the Gutenberg tests are still being recorded, and it looks like this commit is worth investigating further. It was synced to WordPress in r58936. Here's the performance run from that commit in WP-dev.

@joemcgill The code https://github.com/WordPress/wordpress-develop/blob/trunk/src/wp-includes/class-wp-theme-json.php#L3239-L3251 is too expensive see the profiling result. I removed the code and did the comparison.

Server-timing

Mertic	Beta 1	Beta 1 with remove code	Diff %	Diff abs.
Response Time	252.84	247.12	-2.26%	-5.72
wp-before-template	76.26	76.28	0.03%	0.02
wp-before-template-db-queries	3.3	3.35	1.52%	0.05
wp-template	173.42	167.66	-3.32%	-5.76
wp-total	249.87	244.23	-2.26%	-5.64
wp-template-db-queries	2.6	2.59	-0.38%	-0.01

🔴 It clear shows that the code introduce ~3% regression

[mukeshpanchal27]

JFYI, I re-open https://core.trac.wordpress.org/ticket/61858 so editor team also can check it.

[mukeshpanchal27]

Opened PR WordPress/wordpress-develop#7486 that solve the regression. Waiting for the feedback.

[felixarntz]

Since we now reinstated the Code Vitals dashboard, I just took a look at it and noticed https://www.codevitals.run/project/wordpress/home-block-theme-lcpMinusTtfb shows a notable regression that must have happened at some point while it was not working.

It's interesting that our Web Vitals benchmarks for "LCP - TTFB" didn't show a similar regression. Since the Code Vitals dashboard is still using the TT3 theme (which we should probably update at some point), it would be helpful to run another Web Vitals benchmark comparison using TT3, to check whether we see the regression there too - in which case the problem might be something that mostly affects TT3 for some reason.