At a glance the headline numbers for HTTP protocol mix
The HTTP protocol carrying each request: h1, h2, h3.
The HTTP protocol mix mix who uses what, and how fast each group loads
HTTP protocol mix. On the fleet: 61.5% h2, 38.5% h3. 100.0% of sites use at least one h2.
Passing LCP per bucket every category and count level at once - color is the pass rate
Each row is a category, each column its own count bucket (few on the left, many on the right); the cell is the share of those sites passing LCP.
H2 swings the hardest: 86% of sites pass LCP with few, 62% with many. computed
Few vs many - does quantity cost LCP? the pass rate with few vs many of each category
Per category: the pass rate among pages with FEW of it (hollow ring) against pages with MANY (solid dot), worst trend first. Thin buckets are excluded from the endpoints.
More H2 costs the most: the LCP pass rate falls from 86% with few to 62% with many. computed
Why this matters for the Core Web Vitals, and where to start fixing it
HTTP/1.1 gives you a handful of parallel requests per domain and queues the rest. HTTP/2 multiplexes any number of requests over one connection. HTTP/3 keeps that and removes the last stall: on HTTP/2 a single lost packet blocks every stream on the connection, on HTTP/3 only the stream that lost it.
A modern protocol also amplifies consolidation. The fewer domains you use, the more requests share one warm HTTP/3 connection. Requests still travelling over HTTP/1.1 are queueing for no reason.
Chrome field data from 189,915 sites, representing millions of real page loads. How we measured.
Live queries (2) — admin only
SELECT 1 AS ok FROM site_metric_bags LIMIT 0;WITH flat AS (
SELECT b.cat, b.n, b.size, s.crux."all".lcp AS cwv_val
FROM site_metric_bags b
JOIN sites s ON s.origin = b.origin
WHERE b.path = 'network.protocol'
AND b.cat IS NOT NULL
),
totals AS (
SELECT cat,
SUM(n) AS fleet_n,
SUM(size) AS fleet_size,
COUNT(*) FILTER (WHERE n IS NOT NULL) AS sample,
COUNT(*) FILTER (WHERE n > 0) AS with_any,
quantile_disc(n, 0.95) AS p95,
COUNT(*) AS n_rows
FROM flat
GROUP BY cat
),
widths AS (
SELECT *,
GREATEST(1, CAST(round(COALESCE(p95, 0)) AS INTEGER)) AS cap,
LEAST(GREATEST(1, CAST(round(COALESCE(p95, 0)) AS INTEGER)), 12 - 1) AS regular_buckets
FROM totals
),
widths2 AS (
SELECT *,
GREATEST(1, CAST(ceil(cap * 1.0 / regular_buckets) AS INTEGER)) AS width
FROM widths
),
binned AS (
SELECT f.cat, f.cwv_val, w.fleet_n, w.fleet_size, w.sample, w.with_any, w.width, w.regular_buckets,
CASE WHEN f.n IS NULL THEN NULL
ELSE LEAST(CAST(FLOOR(f.n * 1.0 / w.width) AS INTEGER), w.regular_buckets)
END AS bucket_idx
FROM flat f
JOIN widths2 w ON f.cat = w.cat
)
SELECT cat, fleet_n, fleet_size, sample, with_any, width, regular_buckets,
bucket_idx,
COUNT(*) AS n,
quantile_disc(cwv_val, 0.5) AS median,
COUNT(*) FILTER (WHERE cwv_val IS NOT NULL) AS total_cwv,
COUNT(*) FILTER (WHERE cwv_val IS NOT NULL AND cwv_val <= 2500) AS good
FROM binned
WHERE bucket_idx IS NOT NULL
GROUP BY cat, fleet_n, fleet_size, sample, with_any, width, regular_buckets, bucket_idx
ORDER BY cat, bucket_idx;