Router hardware acceleration is slowed down by parental control, and the gigabit phone only runs 400 Mbps

Last month, my childhood friend upgraded our 500MB to gigabit at home, with a plan costing over 30 yuan a month, all for the thrill of downloads. But on the night of the upgrade, the speed test showed 423, which is less than 470 at the original 500 Mbps. The broadband technician came to the site, plugged my laptop straight into the optical modem to test, 940, and said the line was fine, but your router was faulty, so you should switch to a gigabit one. Those two Mercury M6Gs were bought by me at the end of September, gigabit ports—how could they not work? I told him not to pay for now. That night, I remotely checked his backend and found the root cause in ten minutes. This article will break down the matter and explain it clearly.
Let's first eliminate the most easily blamed: it's not Mesh, nor is it network cables
My childhood friend had two M6G Mesh setups. Last year, during my PC build, I wrote an article Mercury Mesh networking, and the study only used wired backlinks. So the first step is to have him plug the laptop into the main router's LAN port for testing, bypassing the wireless and sub-routers. Still 430.
Step two: observe light decay. Their optical modem receives a power of -22 km, within the health line, gigabit port negotiation is also 1000M, and the network cable is fully connected with eight cores. If the physical layer is completely clean, then only the software layer remains. Speed testing methods also have their own rules. Don't use your phone's WiFi to test through a wall. I've written about this process in How to Measure Broadband Speed Accurately. Their laptop is wired for testing, and the data is reliable.
Step three: I remotely melogin.cn to check if any devices were stealing data, but the device list was quiet. But two details stand out: the traffic statistics curve on the backend homepage is open and has never stopped since the day of installation; In device management, his son's tablet had a parental control rule: internet disconnection from 9 p.m. to 7 a.m., set by his wife during the September PC installation period.
What is hardware acceleration? ETC lanes and manual lanes at toll stations
Every data packet passing through the router must be translated: the internal network address of the home device is replaced with a public network address, and the returned packet is claimed by the corresponding device. This process is called NAT, and the translation record serves as a comparison table. Normally, this table is stored in a dedicated forwarding chip for verification, so when a packet arrives, it passes directly, and the CPU rests nearby—this is hardware acceleration: in the TP interface, it's called hardware NAT, overseas versions are called NAT Boost, ASUS is NAT acceleration, OpenWrt is traffic sharing—it's the same thing.
For example, it's a highway toll station. Hardware acceleration is for ETC lanes, allowing cars to pass directly without slowing down, and barrier lifting and lowering rely entirely on sensor devices; CPU soft forwarding is a manual lane, where each car rolls down the window to hand over the card and pay. Even in the manual lane, you can pass quite a few cars in an hour, but once the traffic volume increases, the queue immediately forms. The CPU of a hundred-yuan router is just a toll collector—one person per window, and gigabit bandwidth traffic really can't handle it.
I mentioned this in the article router CPU: flashed OpenWrt but didn't enable data sharing, gigabit dropped to 380MB, CPU soft interrupts occupied 97%, and with the switch ticked, it instantly reverted to 930. My childhood friend's M6G uses the same level of design—4230—the numbers match perfectly.
Where do brands hide this switch
?Each switch has different names and positions, so I looked through them one by one:
| Brand | names, | location, and temperament |
|---|---|---|
| TP-Link/Mercury | Hardware NAT/Hardware Acceleration | New Interface, in the routing settings; After updating the firmware of the Yizhan series, it is enabled by default |
| ASUS | NAT acceleration | ,there is a separate tab with CTF and FA levels, showing the current status |
| Xiaomi/Redmi | rarely appears | , andthe official firmware generally does not provide a switch; by default, it is enabled and you won't be allowed to intervene |
| OpenWrt | traffic loads | network → firewall→ routing settings, divided into software and hardware levels |
According to TP's official statement, some models have it enabled by default and no longer show a switch on the interface. If you can't find it, just pretend it doesn't exist—don't force yourself to look. Xiaomi's one is even more straightforward. If you want to fix this switch, you can only flash OpenWrt first. I wrote about flashing methods in the article OpenWrt flashing.
The real pitfall: once these features are enabled, acceleration quietly disappears
Back to my childhood friend's home. Why is hardware acceleration enabled, still 400 Mbps? Because once parental control is activated, acceleration automatically gives way. The logic is simple: the premise of ETC lanes is that each car doesn't need to be checked individually, but parental control must distinguish who can access each device, traffic statistics must be recorded for each device, and speed limits must monitor each device's bandwidth—all of these require opening the package to read the content, which ETC sensors can't do and must switch all lanes to manual lanes.
This isn't Mercury's fault; it's the rule of the entire industry. TP official FAQ original statement: The QoS feature cannot be enabled simultaneously with NAT acceleration; you must disable NAT acceleration before enabling QoS. ASUS does the same: bandwidth limitation means that when using the NAT, hardware acceleration will automatically turn off. There's an even more subtle story: some ASUS users noticed that NAT acceleration changed from 'enabled' itself to 'handled by the CPU.' After a long investigation, they found out IP traffic monitoring was enabled.
I've listed the features that are easy to disable acceleration in a table:
| Why | does the feature exit upon opening | causesymptoms |
|---|---|---|
| QoS/device speed limits | ; each package must pass classification rules | ; the whole house limit drops by a bit |
| traffic statistics/internet reports | and each device is accounted for | Speed drops, and often no prompts |
| Parental control/internet control | filters by device by time | period. For example, for older |
| models, SPI | firewallsrequire checking the status table | for each packet.Only machines from ten years ago have this issue |
I've also tried throttling for my home TV box. After throttling, I didn't feel laggy because my home is gigabit and the router in the box has a strong CPU. This was written about in the article Mercury speed limit. But when you apply the same operation to a budget phone with control features, it looks like a completely different face. When is QoS worth opening? I separately calculated it in the QoS article >
How do you check if your own acceleration is still active
?
Three methods: from convenience to hardcore work.
ASUS users are the happiest—the NAT acceleration tab directly states the current status, whether it's hardware forwarding or CPU processing—it's decided at a glance. OpenWrt users go to SSH to get a top score. Looking at software interrupt rates, 20% is the acceleration working, and over 80% is the CPU working manual lanes. For regular brands that don't allow viewing, use a simple method: turn off all control functions and test speed once as a baseline; Then open the functions one by one, testing each one. Whenever it dropped more than 20%, the culprit was this one. How speed test numbers are read also matters. Don't take a single web speed test seriously. I wrote this reconciliation logic in the bandwidth and negotiated speed ledgers.
That's how the Childhood Friend Home is positioned: turn off parental controls and traffic statistics, restart, and test speed 910. Then open the parent control separately, 450. Ironclad evidence.
Can you only choose between functionality and speed? Divided into three tiers based on broadband megawattage
Finally, let's talk about what to choose and what to choose. Solution for Sending Friends to Childhood: Parents control the router and move it away; the tablet itself has a youth mode, leaving time management to the device itself, and leaving only clean forwarding on the router. This is the first approach—if you can move control to the device side, don't take up the router's lane.
The second tier is for households with broadband below 500Mbps, so you can easily open it. CPU soft forwarding can run 400 Mbps with no problem; you won't even notice any speed drops. Control and control when needed—don't scare yourself. How many megabits of broadband should be installed to be sufficient? I calculated in the article where many megabytes are installed. For most households, 300 megabits is just a dessert.
The third tier is gigabit broadband, which still requires router-level control. There are two options: either accept dropping to around 400 Mbps, or switch to a machine with a strong enough CPU that won't block the traffic even if it drives in a manual lane—ASUS's Adaptive QoS can coexist with acceleration in some scenarios, but it's expensive. When buying a router, check the parameters first. In the article CPU, I compared three levels to avoid pitfalls.
Finally,here's the order: keep the base value for speed measurement→ find control switches in the backend→ compare each switch against the speed → locate the culprit→ move the device side→ and make trade-offs based on the number of megagrams if it can't. Don't switch routers just because you hear about slow speeds. First, make sure the software costs are clear. A 400-yuan router doesn't necessarily lose to a 1,400-yuan router; it depends on whether you let it focus on its work.
