How to Evaluate Your Website’s Performance with WebPageTest and HTTPArchive Metrics

How to Know Your Website’s Performance?

First, use a widely recognized testing tool. We recommend WebPageTest, an open‑source project from Google’s “make the web faster” initiative.

WebPageTest It originated as an internal AOL tool and was open‑sourced in 2008 under a BSD license. Website: http://www.webpagetest.org/

In this example we test a major site (e.g., Sohu) and capture the main metrics.

After a few minutes the results appear.

We extract the key indicators.

These raise three questions:

What do the metrics mean?

How are the metrics related?

How can I assess my site’s level?

This article analyzes each metric, explains their relationships, and provides a baseline derived from the HTTPArchive database for a more scientific evaluation.

HTTPArchive Database An open‑source repository that records performance data and trends for many websites worldwide, useful for horizontal performance comparisons. Website: http://httparchive.org/

Data used are from the 2015‑03‑15 snapshot.

1. Load Time

Load Time (or onLoad Time) is the interval from navigation start to the window’s load event.

Analysis

This event is widely measured by third‑party tools and is strongly correlated with Visual Complete, Speed Index, and the total number of requests. It also correlates positively with request count: more requests generally mean slower sites.

During the transition from HTTP/1.1 to HTTP/2, measuring Load Time is important because HTTP/2 aims to reuse TCP connections and reduce handshake overhead.

Purpose

Load Time remains a common benchmark for comparing performance across tools such as RUM and WebPageTest, but it is an older metric that does not fully reflect user‑perceived performance. Over time it should be complemented with newer, more relevant indicators.

Value Distribution

All values are in milliseconds.

HTTPArchive data

2. First Byte (TTFB)

Time To First Byte (TTFB) measures the interval from navigation start to the receipt of the first byte of the base page (after redirects).

Analysis

TTFB shows little correlation with other metrics, affecting mainly the start‑render time. It can reveal backend or CDN performance when measuring static resources.

Purpose

TTFB is valuable for assessing CDN efficiency or backend latency and should be part of performance planning.

Value Distribution

All values are in milliseconds.

HTTPArchive data

3. Start Render

Start Render time is the interval from navigation start to the appearance of the first non‑blank content on the screen.

Details are omitted in this article.

4. Visual Complete

Visual Complete attempts to measure the time required to render the above‑the‑fold (ATF) content.

Analysis

Visual Complete is closely related to Fully Loaded, Load Time, and Speed Index. Unless a page contains many lazy‑loaded resources, Visual Complete aligns with Fully Loaded.

Purpose

Because its value correlates with Speed Index and onLoad, measuring Visual Complete alone adds little value, but it is useful when comparing performance before and after lazy loading.

Value Distribution

All values are in milliseconds.

HTTPArchive data

5. Speed Index

Speed Index is a calculated metric that measures how quickly the visible portion of a page is rendered (lower is better).

Analysis

Speed Index is tightly linked to Visual Complete, Start Render, and Load Time, but has low correlation with TTFB and PageSpeed scores.

Purpose

Speed Index reflects content rendering speed, especially for above‑the‑fold elements, and provides a numeric value that is easy to compare.

Value Distribution

For content‑heavy sites, an ideal target is around 1000.

Values are expressed in units, not time.

HTTPArchive data

6. Total Number of Requests

This metric counts how many requests a page makes to the server before loading completes.

Analysis

Request count correlates with third‑party domains but also shows strong relationships with Fully Loaded, Visual Complete, and onLoad.

Purpose

If a site relies heavily on third‑party tags, a high request count can indicate performance degradation, making this metric useful for monitoring third‑party impact.

Value Distribution

Values are expressed in request units.

HTTPArchive data

7. PageSpeed Insights

Google PageSpeed measures network‑independent performance factors such as server configuration, HTML structure, and the use of images, JavaScript, and CSS.

It provides separate scores for mobile and desktop, ranging from 0 to 100; scores above 85 indicate good performance.

Analysis

PageSpeed has very low correlation with other metrics, highlighting its independence. It is negatively correlated with time‑based metrics: lower times yield higher scores.

Purpose

PageSpeed identifies structural issues (e.g., render‑blocking JavaScript or CSS) that other metrics may miss, making it an essential part of a performance‑planning toolkit.

Value Distribution

Values range from 0 to 100.

HTTPArchive data

8. Total Bytes

Total Bytes represent the sum of all downloaded objects from the first byte received until the page finishes loading.

Analysis

Total Bytes have a strong correlation with Fully Loaded, Visual Complete, and Load Time, though the relationship is non‑linear.

Purpose

This metric helps detect sudden size growth caused by large images or new JavaScript libraries, providing a useful overall health indicator.

Value Distribution

Values are expressed in bytes.

HTTPArchive data

9. Number of Domains

This metric counts the distinct domains from which a page loads resources.

Analysis

More domains usually indicate a busier site and slightly higher Fully Loaded times, though the correlation is modest.

Purpose

Tracking domain count helps monitor third‑party fragmentation; limiting domains and deferring third‑party loads can reduce their impact on perceived performance.

Value Distribution

Values are expressed as a count.

HTTPArchive data

Performance Metrics Summary

HTTPArchive’s extensive desktop‑site dataset shows that Speed Index, Load Time, and PageSpeed scores have clear relationships with perceived performance. Metrics such as domain count, request count, and DOM element count also correlate with Speed Index.

When measurement resources are limited, prioritize Speed Index, Load Time, and PageSpeed scores. If third‑party pressure is high, also track domain count and total request count to provide data‑driven guidance to business owners.

Testing Metric Correlations

We used the HTTPArchive database, extracting non‑null values and calculating Pearson and Spearman correlation coefficients. Coefficients above ±0.7 were considered significant; those below ±0.4 were deemed weak.

Metrics that stood out for their richness and complementary perspectives include Speed Index, Load Time, Google PageSpeed, TTFB, and total domain count.