Recently, I gradually separated three access entry points of my WordPress site from Tencent Cloud EdgeOne:
media.shuijingwanwq.com
en.shuijingwanwq.com
admin.shuijingwanwq.com
The current structure is roughly as follows:
www.shuijingwanwq.com
→ EdgeOne
→ Chinese site frontend
en.shuijingwanwq.com
→ Cloudflare
→ English site frontend
media.shuijingwanwq.com
→ Cloudflare
→ Images and attachments
admin.shuijingwanwq.com
→ Cloudflare DNS Only
→ Alibaba Cloud Hangzhou origin server
Previously, I analyzed this separation from a cost perspective. Based on the traffic estimates at the time, the actual monthly cost of EdgeOne was expected to drop by more than half after the separation.
However, after reducing costs, whether the website’s current access performance remains acceptable is the question that needs to be confirmed more urgently.
Therefore, this time I did not focus on a strict comparison before and after the separation, nor did I restore the old architecture for testing. Instead, I directly verified the actual performance in the current production environment:
- Whether access to the Chinese site from users in mainland China is still fast enough;
- How access to the English site and media resources performs for users in mainland China;
- Whether access to the Chinese and English sites is stable for overseas users;
- Whether migrating images to Cloudflare has noticeably slowed down the first-screen experience;
- Whether the current architecture is worth keeping.
1. Test Subjects
To minimize interference from differences in page content, I selected a pair of Chinese and English articles that are translations of each other.
Chinese article:
https://www.shuijingwanwq.com/2026/07/11/19354/
English article:
https://en.shuijingwanwq.com/2026/07/11/19362/
Both articles use the same featured image:
https://media.shuijingwanwq.com/2026/07/1-36.png
This set of pages is well-suited as test subjects for the following reasons:
- The Chinese and English article contents basically correspond;
- The two HTML pages are almost identical in size;
- Both articles use the same featured image;
- The performance of EdgeOne and Cloudflare when returning HTML can be observed separately;
- The image loading speed of the
mediadomain can be tested independently.
Local download results show:
Chinese page: 456368 bytes
English page: 455941 bytes
Featured image: 339881 bytes
The size difference between the Chinese and English HTML pages is only about 427 bytes, so any significant performance differences observed later cannot simply be attributed to different page sizes.
2. Confirming Cache Status Before Testing
Before testing, I continuously requested the three URLs via curl to confirm that the cache had reached a normal state.
First response for the English article:
cf-cache-status: EXPIRED
From the second request onwards, it changed to:
cf-cache-status: HIT
age: 26
First response for the media image:
cf-cache-status: MISS
From the second request onwards, it changed to:
cf-cache-status: HIT
age: 26
The Chinese article did not show a Cloudflare-style HIT, but age continued to increase:
age: 27
age: 56
age: 125
This indicates that the EdgeOne node was continuously returning the same cached content.
Therefore, all three test subjects met the cache hit conditions during the official speed tests:
Chinese article: EdgeOne cache valid
English article: Cloudflare HIT
Featured image: Cloudflare HIT
3. Testing Methods
This time, I used three testing methods in total.
1. BOCE Mainland China Nodes
Used to observe:
- China Telecom;
- China Mobile;
- China Unicom;
- Average response across regions;
- Timeouts or inaccessible nodes.
Since the first BOCE test often encounters unestablished node caches or slow line initialization, each URL was tested twice consecutively, primarily recording the second result.
2. BOCE Overseas Nodes
The same platform was used to test multiple overseas regions, avoiding inconsistent testing standards across different tools.
Again, tested twice consecutively, using the second result as the primary reference.
3. WebPageTest
Using the following configuration:
Device: Desktop - Chrome
Network: Cable
Runs: 3
Repeat View: Disabled
Test regions selected:
Singapore
Dulles, Virginia, USA
Primary observations:
- TTFB;
- FCP;
- LCP;
- Speed Index;
- Fluctuations across the three runs.
Additionally, on my own Chengdu Mobile home broadband, I continuously accessed the Chinese and English pages and images via curl to confirm the performance on a real local network.
4. BOCE Mainland China Test Results
1. Chinese Article: EdgeOne Performance is Very Stable
The second BOCE test result for the Chinese article:
| Item | Result |
|---|---|
| National average response | 0.307 seconds |
| Telecom average response | 0.299 seconds |
| Mobile average response | 0.418 seconds |
| Unicom average response | 0.197 seconds |
| Inaccessible nodes | 1 |
The vast majority of regions on the map are in the green zone.
From these results, keeping the Chinese site HTML on EdgeOne was the right decision.
Even though the Chinese article page itself is nearly 456KB, after a cache hit, the national average response is still only about 0.3 seconds, and there is no particularly severe disparity among the three major carriers.
2. English Article: Cloudflare is Noticeably Slower in Mainland China
The second BOCE test result for the English article:
| Item | Result |
|---|---|
| National average response | 1.855 seconds |
| Telecom average response | 1.532 seconds |
| Mobile average response | 1.561 seconds |
| Unicom average response | 2.473 seconds |
| Inaccessible nodes | 4 |
Compared to the Chinese article:
Chinese article: 0.307 seconds
English article: 1.855 seconds
The average response for the English article is about 6 times that of the Chinese article.
In particular, some Unicom and southwestern region nodes have noticeably high response times, even reaching 10 seconds on some nodes.
This indicates that after migrating the English site to Cloudflare, access performance from mainland China did indeed suffer a noticeable decline.
However, the English site is primarily aimed at overseas users in the future, so this is an expected trade-off.
3. media Images: Loading Speed in Mainland China is Also Slow
The second BOCE test result for the featured image:
| Item | Result |
|---|---|
| National average response | 1.514 seconds |
| Telecom average response | 1.124 seconds |
| Mobile average response | 1.618 seconds |
| Unicom average response | 1.846 seconds |
| Inaccessible nodes | 6 |
This image is only about 332KB, but the average response still exceeds 1.5 seconds.
From the results, the impact of separating media from EdgeOne is more noticeable for mainland China users than for the Chinese HTML.
Now, when opening a Chinese article, the actual link path becomes:
HTML, CSS, JavaScript
→ EdgeOne
Featured image, in-content images
→ Cloudflare
Therefore, the text content of the Chinese article can appear very quickly, but the featured image and in-content images may load slightly later.
5. Chengdu Mobile Home Broadband Real-World Test
BOCE can observe nodes nationwide, but my own daily network is Chengdu Mobile, so I also conducted continuous request tests locally.
1. English Article via Cloudflare
Three consecutive requests:
1st time:
HTTP=200
TTFB=3.346020s
TOTAL=6.057315s
2nd time:
HTTP=200
TTFB=1.399212s
TOTAL=4.266293s
3rd time:
HTTP=000
TOTAL=16.376829s
OpenSSL unexpected eof while reading
Although the first two requests returned successfully, the complete downloads took about 6.06 seconds and 4.27 seconds, respectively.
The third request encountered an SSL connection exception after waiting about 16 seconds, and the page did not return successfully.
2. Chinese Article via EdgeOne
Same computer, same Chengdu Mobile broadband, same testing method:
1st time:
HTTP=200
TTFB=0.103798s
TOTAL=0.224673s
2nd time:
HTTP=200
TTFB=0.094631s
TOTAL=0.148890s
3rd time:
HTTP=200
TTFB=0.069811s
TOTAL=0.119547s
All three were successful, and the complete download times were all under 0.23 seconds.
This comparison is very intuitive:
| Page | CDN | Page Size | Complete Download |
|---|---|---|---|
| Chinese article | EdgeOne | 456368 bytes | 0.120～0.225 seconds |
| English article | Cloudflare | 455941 bytes | 4.266～6.057 seconds |
| English article 3rd time | Cloudflare | 455941 bytes | Connection failed |
The sizes of the two pages are almost identical, but the access results differ vastly.
Therefore, at least on my Chengdu Mobile broadband, a clear conclusion can be drawn:
EdgeOne returns the Chinese page very quickly and stably; Cloudflare returns the English page noticeably slower, with a risk of connection interruptions.
It must be emphasized that this is only the real-world test result of the local Chengdu Mobile link, and cannot directly represent all Mobile and Unicom users nationwide.
However, it is basically consistent with the phenomenon of slower responses on some Mobile and Unicom nodes in BOCE.
3. Chengdu Mobile Accessing media Images
Three consecutive requests for the same featured image:
1st time:
TTFB=0.815904s
TOTAL=2.872365s
SPEED=118327 bytes/s
2nd time:
TTFB=2.907905s
TOTAL=5.342874s
SPEED=63613 bytes/s
3rd time:
TOTAL=16.225433s
OpenSSL unexpected eof while reading
The image is only 339881 bytes, but the second download speed was only about 62KB/s, and the third again encountered an SSL connection exception.
This indicates that the issue of Chengdu Mobile accessing Cloudflare affects not only the English HTML but also media images.
6. BOCE Overseas Node Testing
1. Chinese Article Overseas Performance
Second overseas test result for the Chinese article:
| Region | Average Response |
|---|---|
| Global average | 0.802 seconds |
| Asia | 0.677 seconds |
| North America | 0.751 seconds |
| South America | 0.749 seconds |
| Europe | 0.563 seconds |
| Africa | 1.287 seconds |
| Oceania | 0.690 seconds |
This result is not entirely consistent with the initial theoretical expectation.
Previously, I thought EdgeOne might be noticeably weaker than Cloudflare overseas. But from these actual results, the performance of the Chinese article on overseas nodes is not bad, with a global average response of only about 0.8 seconds.
In particular, the average response in Europe is about 0.563 seconds, and in Oceania about 0.690 seconds.
Therefore, there is currently no evidence that:
EdgeOne is definitely noticeably slower than Cloudflare overseas
At least for the current Chinese article, the current cache state, and the BOCE overseas nodes, EdgeOne’s performance is acceptable.
2. English Article Overseas Performance
Second overseas test result for the English article:
| Region | Average Response |
|---|---|
| Global average | 1.347 seconds |
| Asia | 1.479 seconds |
| North America | 1.302 seconds |
| South America | 0.916 seconds |
| Europe | 0.863 seconds |
| Africa | 2.154 seconds |
| Oceania | 1.175 seconds |
This result is noticeably more normal than the 6.314 seconds from the first test.
From the second result, Cloudflare performs well overseas overall, but it is not noticeably faster than the EdgeOne Chinese site in BOCE.
Chinese article EdgeOne: 0.802 seconds
English article Cloudflare: 1.347 seconds
Of course, although the two pages are similar in size, their domains, cache nodes, and page requests are not exactly the same, so this set of numbers cannot be taken as a strict CDN performance ranking.
A more accurate conclusion is:
The current Chinese site via EdgeOne does not have obvious performance issues for overseas access; the English site via Cloudflare also performs within an acceptable range overseas.
3. media Image Overseas Performance is Not Stable Enough
Second overseas test result for the media image:
| Region | Average Response |
|---|---|
| Global average | 3.717 seconds |
| Asia | 4.853 seconds |
| North America | 4.030 seconds |
| South America | 3.000 seconds |
| Europe | 1.051 seconds |
| Africa | 4.951 seconds |
| Oceania | 5.760 seconds |
This set of results is noticeably weaker than the two HTML pages.
However, the image test includes the complete file download time, whereas the “response time” in the HTML page test is not exactly the same as the complete page rendering metric.
Therefore, this chart is more suitable for illustrating:
Media resources on Cloudflare fluctuate significantly across different overseas nodes.
Whether it truly affects the user’s first-screen experience still needs to be judged in combination with the LCP results from WebPageTest.
7. WebPageTest Singapore Testing
1. Chinese Article
Under the Singapore node, the Chinese article underwent three First View tests.
One of the results had incomplete data, so I primarily referenced the other two valid results.
The valid result range is:
| Metric | Result |
|---|---|
| FCP | 1.256～1.776 seconds |
| LCP | 1.501～2.048 seconds |
| Speed Index | 1.585～2.032 seconds |
| CLS | 0.003 |
Although the test page showed Page Load Timeout, the first-screen content actually displayed very quickly.
Upon further checking the request list, I found that the long-tailed requests mainly came from:
fundingchoicesmessages.google.com
pagead2.googlesyndication.com
n.clarity.ms
Which are Google Funding Choices, Google Ads, and Microsoft Clarity.
Therefore, Page Load Timeout is not a first-screen loading failure caused by the Chinese HTML, EdgeOne, or media images, but rather third-party advertising and analytics scripts continuing to make requests after the main page content has finished loading.
2. English Article
The Singapore test results for the English article were very stable:
| Metric | Median Result |
|---|---|
| TTFB | 0.194 seconds |
| FCP | 1.264 seconds |
| LCP | 1.478 seconds |
| Speed Index | 1.703 seconds |
| Total Blocking Time | 0.180 seconds |
| CLS | 0 |
The three LCP values were:
1.801 seconds
1.487 seconds
1.478 seconds
This indicates that when accessing the Cloudflare English site from Singapore:
- HTML first byte is very fast;
- First-screen content appears in about 1.3 seconds;
- The largest content element finishes loading in about 1.5 seconds;
- Fluctuation across the three runs is very small.
Therefore, although Cloudflare performs poorly when accessed via Chengdu Mobile, the first-screen performance of the English site under the Singapore node is very good.
8. WebPageTest Virginia, USA Testing
1. Chinese Article
The median result for the Chinese article in Virginia, USA was:
| Metric | Result |
|---|---|
| TTFB | 0.217 seconds |
| FCP | 0.826 seconds |
| LCP | 1.214 seconds |
| Speed Index | 1.415 seconds |
| Total Blocking Time | 0.223 seconds |
| CLS | 0.003 |
The three LCP values were:
1.149 seconds
1.214 seconds
1.193 seconds
There was almost no fluctuation across the three results.
This again shows that the current EdgeOne Chinese site is not slow when accessed from the USA.
In particular:
FCP: 0.826 seconds
LCP: 1.214 seconds
This is already considered very good first-screen performance.
2. English Article
The median result for the English article in Virginia, USA was:
| Metric | Result |
|---|---|
| TTFB | 0.178 seconds |
| FCP | 1.594 seconds |
| LCP | 1.838 seconds |
| Speed Index | 1.780 seconds |
| Total Blocking Time | 0.106 seconds |
| CLS | 0 |
The three LCP values were:
3.052 seconds
1.459 seconds
1.838 seconds
The first run was slow, but the latter two recovered noticeably.
Overall, the first-screen performance of the English site in the USA remains at a good level, though the fluctuation is slightly larger than that of the Chinese site.
Notably, the Chinese article under the US node was actually faster than the English article:
| Page | FCP | LCP | Speed Index |
|---|---|---|---|
| Chinese article EdgeOne | 0.826 seconds | 1.214 seconds | 1.415 seconds |
| English article Cloudflare | 1.594 seconds | 1.838 seconds | 1.780 seconds |
This shows that, at least in this test, one cannot simply assume:
Cloudflare is definitely faster than EdgeOne overseas
The actual performance of a CDN is also affected by:
- Current node;
- Page cache;
- Routing;
- Page resources;
- Third-party scripts;
- Testing period;
and other factors.
9. Actual Performance Conclusions Under the Current Architecture
Combining BOCE, Chengdu Mobile local tests, and WebPageTest, the following conclusions can be drawn.
1. Accessing the Chinese Site from Mainland China: Very Good Performance
The Chinese HTML continues to use EdgeOne, and its current performance is very stable.
BOCE national average response:
0.307 seconds
Chengdu Mobile local complete download:
0.120～0.225 seconds
This shows that the current architecture retains the most critical performance foundation:
The Chinese HTML, primarily aimed at mainland China users, still uses EdgeOne, which is more suitable for mainland access.
2. Accessing the English Site from Mainland China: Noticeably Slower
After switching the English HTML to Cloudflare, the access performance from mainland China is noticeably weaker than the Chinese site.
BOCE average response:
1.855 seconds
Chengdu Mobile local complete download:
4.266～6.057 seconds
Moreover, an SSL connection exception occurred during consecutive requests.
Therefore, mainland China users accessing the English site is the scenario with the most noticeable performance cost in the current architecture.
However, the primary target audience for the English site is overseas readers, so this trade-off remains acceptable for now.
3. Accessing Chinese Site Images from Mainland China: HTML is Fast, but Images May Be Slightly Slower
The HTML for the Chinese article is very fast, but the featured image and in-content images use Cloudflare.
BOCE average response for the media image:
1.514 seconds
When downloading a ~332KB image locally on Chengdu Mobile:
2.87～5.34 seconds
And one connection failure occurred.
Therefore, when mainland China users open a Chinese article, the actual experience might be:
Title and body text appear very quickly
Featured image displays slightly later
In-content images load when scrolled near
Since WordPress has already enabled lazy loading for images, images below the first screen will not all block the initial page, so this issue is not yet severe enough to require an immediate architecture change.
4. Accessing the Chinese Site from Overseas: EdgeOne is Not Slow
Whether in BOCE or WebPageTest, the overseas performance of the Chinese site is good.
Virginia, USA:
FCP: 0.826 seconds
LCP: 1.214 seconds
Singapore:
LCP: 1.50～2.05 seconds
Therefore, the previous theoretical judgment that “overseas users accessing the Chinese site might be noticeably slowed down due to EdgeOne” was not supported in this test.
5. Accessing the English Site from Overseas: Cloudflare Performs Well
Singapore English site:
TTFB: 0.194 seconds
FCP: 1.264 seconds
LCP: 1.478 seconds
USA English site:
TTFB: 0.178 seconds
FCP: 1.594 seconds
LCP: 1.838 seconds
This shows that Cloudflare can effectively handle the overseas English site.
Although individual runs still show some fluctuation, the overall first-screen performance is at a good level.
6. media is the Part of the Current Architecture That Needs Continued Observation
The English HTML performs very well in overseas WebPageTest, but the media images showed significant fluctuation in both BOCE domestic and overseas nodes, as well as in the Chengdu Mobile local test.
Therefore, the part of the current architecture most worth long-term observation is not the English HTML, but:
media.shuijingwanwq.com
Going forward, I can focus on observing:
- Whether the featured image is frequently delayed;
- Whether the LCP is primarily composed of the media image;
- Whether China Mobile and Unicom access remains consistently unstable;
- Whether further compression of the featured image is needed;
- Whether it is necessary to prioritize generating WebP or AVIF;
- Whether the size of the first-screen images in articles needs to be reduced.
10. Comparison Before and After Separation is No Longer the Focus of This Article
Strictly speaking, this test did not restore the complete architecture from before the separation, so it cannot be called an A/B test under completely identical conditions.
Previous historical data can only serve as background reference:
- EdgeOne is usually more stable across mainland China nodes;
- Cloudflare Free has significant fluctuations on some Mobile and Unicom lines;
- After separating the English site and media resources, EdgeOne costs dropped significantly;
- After separating admin, WordPress long requests are no longer affected by CDN timeouts.
What this article truly answers is:
After completing the separation, is the actual performance of the current production environment acceptable?
From the test results, the answer is yes, but there are clear regional differences.
11. Whether the Current Architecture Needs to be Readjusted
There is currently no need to immediately overturn the existing plan.
Chinese Site Continues to Keep EdgeOne
This is the clearest conclusion at present.
The Chinese site is the primary source of traffic, access speed in mainland China is very good, and overseas access also has no obvious shortcomings.
English Site Continues to Keep Cloudflare
The English site is slower to access from mainland China, but performs very well in WebPageTest in Singapore and the USA.
Considering that the English site is primarily aimed at overseas users, continuing to keep Cloudflare is reasonable.
media Temporarily Keeps Cloudflare, but Continues to be Observed
Migrating media resources off EdgeOne is one of the main sources of cost reduction.
Although there is some performance loss when accessing images from mainland China, the Chinese HTML and first-screen text are still very fast, and the LCP in overseas WebPageTest is also within a good range.
Therefore, for the time being, it can be kept as is, and there is no need to migrate back to EdgeOne just because of a single test.
Going forward, it is more appropriate to start with image size and format optimization, rather than immediately changing the CDN architecture.
admin Does Not Use Public Access Speed as Its Primary Metric
admin.shuijingwanwq.com uses DNS Only to directly connect to the Hangzhou origin server.
Its primary goals are:
- Avoiding CDN long request timeouts;
- Ensuring stable WordPress upgrades;
- Ensuring stable plugin installations and translation requests;
- Reducing backend traffic going through EdgeOne.
Therefore, the backend separation is primarily a stability optimization, not a global access acceleration.
12. Final Conclusion
This test shows that after separating media, en, and admin from EdgeOne, the current website has formed very distinct layered access characteristics.
Mainland China accessing the Chinese site:
Very fast, EdgeOne performance is stable
Mainland China accessing the English site:
Noticeably slower, Cloudflare link experiences fluctuations
Mainland China accessing media:
Image loading is slow, especially noticeable on Mobile lines
Overseas accessing the Chinese site:
Good performance, EdgeOne shows no obvious shortcomings
Overseas accessing the English site:
Good performance, Cloudflare first-screen speed is stable
WordPress backend:
Focuses on long request stability, not global access speed
Therefore, this separation did not make all scenarios faster.
What it truly achieved is:
Retaining the core performance of the main Chinese site in mainland China, transferring the English site and high-traffic media resources to a platform more suited to the current cost structure, and simultaneously bypassing the CDN for the WordPress backend in exchange for a more stable management and upgrade experience.
From the current results:
- The performance of the main Chinese site has not been noticeably affected;
- The overseas performance of the English site is good;
- EdgeOne costs have dropped significantly;
- Backend long requests are more stable;
- The main trade-off is that access to the English site and media images from mainland China is noticeably slower.
For a website that currently focuses primarily on mainland China Chinese users while gradually developing overseas English content, this trade-off remains reasonable.
At this stage, there is no need to readjust the overall architecture again.
What truly needs continuous attention is the stability of Cloudflare on China Mobile and Unicom lines, and whether media images will consistently slow down the LCP in real-world access.
Related Records
- Is Cloudflare Free suitable for a mainland China WordPress main site? A real-world review from cache HIT to gray cloud direct connect to Alibaba Cloud ECS
- How much faster is it after migrating from Cloudflare Free to EdgeOne? A three-stage WordPress CDN BOCE and WebPageTest comparison
- EdgeOne traffic cost troubleshooting: Migrating WordPress uploads static resources to a Cloudflare media subdomain
- Migrating the WordPress backend to an independent admin subdomain: Bypassing EdgeOne to resolve core upgrade 524 timeouts
- After separating media, en, and admin from EdgeOne, I re-estimated the website’s actual monthly CDN costs
需要长期技术维护或远程问题排查？
我是拥有 15+ 年经验的 PHP / Go 后端工程师，长期关注已有系统维护、Bug 修复、性能优化、服务器排查、WordPress 网站维护和小功能迭代。
如果你的项目遇到以下情况，可以先从一次小问题排查开始合作：
- ✅ PHP / Laravel / Yii2 老项目无人维护
- ✅ Go / Gin 后端接口需要排查或优化
- ✅ WordPress 网站访问慢、报错或插件冲突
- ✅ Nginx / MySQL / Redis / Linux 服务器异常
- ✅ CDN / Cloudflare / DNS / HTTPS 配置问题
- ✅ 需要长期远程技术支持或兼职维护
更多介绍请查看：关于我 & 合作
微信：13980074657
邮箱：shuijingwanwq@gmail.com
Telegram：@shuijingwan
GitHub：https://github.com/shuijingwan
发表回复