What does CDN mean? Speed tests show 940 Mbps and 4K speeds while spinning. I checked which server the video actually came from

speed tests can reach 940 Mbps, but the living room TV is spinning in 4K online, making people nervous. What's even more infuriating is that the same movie, when you turn on your phone and watch it, it plays in an instant.
Last time I wrote about the TV network, I left a tail note: the network cable is fine, but during rush hour, it gets . This is definitely not because of the home network. This article breaks down the tail—the video you watch, which server is it from?
video isn't sent from headquarters; it's taken from a branch downstairs at your home. The
CDN three letters are called the Content Delivery Network. What they do can be summed up in one sentence: the website copies its content in advance and sends it to data centers across the country. When you watch the video, you pick up the closest copy to you.
to give an example, a milk tea chain. The main store is in Guangdong, so if you want to drink in Beijing, you don't have to wait for Guangdong to send you a cup—the downstairs branch serves the same cup. Video websites are the same; there can't be just one data center; popular dramas and videos are stored in dozens of data centers nationwide. These data centers have a unified name: edge nodes; The action of going to the headquarters computer room to adjust scores is called 'returning to the source.' What would
be like without a CDN? With hundreds of millions of users nationwide crammed into one data center, no matter how wide the bandwidth is, it's still not enough to divide the data. Even more critical is the physical distance: you're in Harbin, the servers are in Guangzhou, and just running data back and forth takes tens of milliseconds. The video is done in segments, and the debt that runs away has to be paid repeatedly.
branches also have special considerations for what to stock. Popular dramas are something everyone wants; each branch stocks up in advance so you can get them when you arrive; Niche old videos rarely get watched, and most branches don't stock them. If you open them once, you have to go back to headquarters for fresh adjustment—that's why popular content opens instantly, and some niche videos are always a beat slow in the first few seconds, having nothing to do with your network.
so to check if the video card is lagging, the first thing is not to check the router, but to find out which branch you were sent to.
how does it know which one to give you? No matter how many branches there are in the DNS
you pass when you ask for directions, you have to decide first which one to send you to. The CDN relies on the DNS you pass through when you ask for directions. DNS analyzed the process of recursive queries the CDN scheduling is hidden in this step.
process breaks down like this: you enter the website, your DNS goes to Bilibili's dispatch system; The dispatch system doesn't rush to answer; it first glances at "Who asked this question for you"—that is, where your DNS exit is, which carrier you are passing through, then picks a batch of IPs from the node pool that best fit you and throws them back.
I tested it on my computer. Ask for Bilibili's domain name, and you can get back 16 IPs at once, all from the same pool; Ask Taobao and get back two IPs; Ask a few more times, and the order of the IPs will change. This isn't hijacking—it's CDN taking turns distributing work in the pool, just like several stalls at the market taking turns weighing the weight.
pitfalls lie in the scheduling reference: it looks at DNS locations, not your location. If you use Unicom's broadband but the DNS outlet is on the telecom side, you might be sent to the telecom node, causing data to loop across the network in a long loop. This is called inaccurate scheduling and is one of the hidden reasons for checking video cards.
industry has a patch called ECS, short for EDNS Client Subnet. When DNS asks for directions, it also includes your IP prefix, allowing the dispatch system to get your real approximate location and dispatch stores accurately. Alibaba 223.5.5.5, Tencent 119.29.29.29, Baidu 180.76.76.76 all support public DNS. So when people say online that 'switching DNS makes watching videos faster,' it's mostly not that the DNS itself runs faster, but that the CDN has switched to a closer branch.
I went through them one by one: the nearest node is ten times the difference from the headquarters data center
just talking about it without practicing is pointless. I tested the TCP handshake latency one by one on the IP string I retrieved—that is, the round-trip time of the three handshakes established for the connection. The meaning is similar to ping. The article on what latency is explained how to read this number. The results are recorded in the table below.
| connecting | handshake and round trip? What does it feel like? | |
| B station is the nearest node | 18ms | local branch |
| B Station Nearby Node 2 | 21ms | another local |
| B node slightly farther away | 35ms | the branch in neighboring cities |
| Taobao node | 18ms | Same-city branch |
| comparison: A small site without CDN | 209ms | direct mail from headquarters |
the nearest node is ten times worse than the source. What does 209ms mean? Every time a connection is established, just running takes 0.2 seconds. A webpage requires loading over a dozen files, and videos need to be streamed in segments—all of this is tied to the system.
you can verify this difference at home yourself: type the domain name on nslookup on your computer, get a string of IPs, then find a multi-location ping-based website to test it. You'll see latency measured by the same domain in different provinces ranging from tens of milliseconds to over a hundred milliseconds—that is, the distance difference between branches in different regions, crystal clear.
evening rush hour, three suspects lined up one by one
understand the CDN, and the "speed check video card" issue can be triaged. The speed testing software selects the nearest node and runs a single task. speed test article said this is an ideal ; Watching videos is in the real world; nodes require queuing, cross-networking, and possibly even a return to the source. Evening rush hour, three suspects in total.
| suspect | typical characteristics | who will handle the crowding |
| CDN Crowding: At 8 p.m., tens of thousands crowded into the same branch | smooth during the day, It lags during evening rush hour, lowering the resolution by one level immediately makes | the platform's decision. You have to stagger or downgrade your speed |
| cross-network detours: Unicom broadband is sent to telecom nodes | speed tests are normal, but certain websites are slow to | Try switching to a public DNS service that supports ECS |
| your own network segment: poor WiFi signal or bandwidth occupied by family members | slow face-to-face test on phone, slow website testing | try it yourself, first plug in the network cable to eliminate wireless bands | __ The HTML133__
order is recommended from bottom to top: first plug in the network cable and test it face-to-face, then clean the wireless segment—the 2.4G congestion and wall have calculated the frequency band, so your home signal blind spots can also be checked and measured by the meter; The network cable was still delayed, lost packets were ruled out, and was ruled out, so it was the two suspects outside the wall who were finally involved.
mobile data is actually smoother, making it the most useful triage signal here. Traffic goes through the operator's own core network and local scheduling, which means it bypasses your home broadband entire chain—it is good for your lag, but the problem lies in your home broadband or scheduling; It was also lagging, so it was the platform's nighttime milestone. Time to sleep.
three things you can do at home, and the one that really works with you
is checking DNS. Check both the DNS on the router's WAN port and the DNS sent to the device by DHCP. DNS configuration article have the backend paths from each provider. To put it counterintuitively: Home broadband users use the ISP's default DNS, so scheduling is actually the most accurate tier, because DNS is on the same network as you; If you want to switch to public DNS, go for those companies that support ECS mentioned earlier. Don't just randomly fill in the 'speed DNS boost' recommendations.
the second active option, test your own segment completely before drawing conclusions. Conduct a face-to-face test with wired bandwidth, and note down each time slot— bandwidth, negotiation, and actual testing are the three of the bill. Don't blame broadband on the numbers from the WiFi wall.
can move, report the repair, bring evidence. "Watch 4K cards from 8 to 10 p.m., no buffering during the day, normal mobile data, direct network connection speed test 940"—these few sentences were reported, and the technician immediately knew whether to check the line or the bureau. Before reporting, I also checked the light decay reading myself. Only within -27 can I feel confident. the light decay article have complete segmentation.
as for platform node crowding, niche content retrieval, and cross-network congestion, all three are outside your home and can't be fixed by anyone. Lowering the resolution by one stop and watching at different peaks are the two most practical tricks.
By the way, some DNS settings in TV boxes and smart TVs are built-in and simply can't be changed in the settings. No need to argue with it—just change the DNS sent by DHCP on the router. The device still has to go through the home router station to get directions, which means you have a different guide for the router.
wrapping up with a clear order: face to face test excluding your own server—see if it's full-time or split-time lag (checking DNS and line during all times, as time periods often cause node crowding)—try switching to a DNS that supports ECS—if that doesn't work, bring the time period record and light decay readings and request repair. After completing the four steps, this was a matter of self-respect.
