What happens if the router reaches full connections? The night I logged into BT, the webpage couldn't open at all. I've figured out the NAT session list account

last month, I added a batch of preservation tasks to the NAS download device, and the list grew from about thirty to over two hundred. At 10 p.m. on the third day, my wife called out to me in the bedroom: 'My phone and web pages are all spinning, and I can't find them even when casting to TV.' I checked her phone and saw that the problem was that WeChat messages kept popping up one after another, Channels wouldn't work, and web pages wouldn't open.

can receive messages, it means the network hasn't been disconnected—it's 'new connections can't be built.' Restarting the router immediately restored FamilyMart. I thought it was over, but around 10 p.m. the next night, the exact same symptoms came back again. This pattern of "reactivation at the right time, restarting and reviving" led me to something most people have never heard of—the router's NAT session table.

router has a connection ledger that is full and refuses customers

Every device at home goes online not just by a single wire, but by connecting one by one. Scrolling a web page on your phone is backed by a DNS page, a web server, images, and several ad domains. When the router forwards NAT data, every connection must be recorded in memory: whose internal network IP it is, which port it comes from, which IP and port it connects to, and whether TCP or UDP is used. This record is called a NAT session, also known as a connection tracking entry, and in Linux systems it's called a conntrack.

this ledger has two deadly attributes. First, page count is limited the router's memory and manufacturer provide it are determined by the manufacturer. A single record occupies about 300 bytes of memory, which may not sound like much, but for a home computer with over 200MB of memory, after deducting the system space and the ledger space left for the ledger, it generally amounts to just over a few thousand to just over ten thousand yuan. ASUS's mid-to-high-end models are quite generous, with the spec page directly showing 16,384 or 32,768 NAT sessions; Phones costing several hundred yuan don't even list this amount; even if you ask customer service, they won't get an answer. Second, an account isn't immediately closed once the connection is disconnected After a TCP connection breaks, entries must be cleared from the table by timeout timeout. For connections like UDP, sending packets is recorded and gradually expires due to timeouts. Once the car leaves, you still have to wait for the management staff to display the license plate—that's what it means.

the moment the ledger was full, the router's action was straightforward: new connections, forgotten, packet loss. So the old connections that have already been established continue to live (WeChat's long connections survived this way), and all requests to create new connections are completely lost. This also explains why restarting the router works so well—restarting clears the entire ledger, frees up all parking spaces, and of course, restores immediately.

why is it easiest to overload the ledger with a BT connection

In my home under normal conditions, during the evening rush hour, 40 to 50 devices are online, only about 1,500 to 600 connections. I later saw this number on a soft router and matched I wrote about memory before. The problem lies in breed conservation.

BT connection model is inherently connected by black holes. A seed does not connect to one server; it connects dozens of peers simultaneously and takes a small piece each; Over two hundred seeds are planted together, and the connections are built upward. To make the upload volume look better, I even changed the global maximum connection count in qBittorrent from the default 500 to 2000, and also relaxed the single seed cap—just adding fuel to the fire. Additionally, there are two hidden ledgers: one is DHT, where the client maintains a routing table for hundreds or thousands of nodes, and every UDP communication between nodes is recorded in the ledger; Second, each seed's tracker had to regularly report progress. Over two hundred seeds each had two or three trackers, and hundreds of short conversations were repeatedly reviewed. Disconnected connections on the TCP side still need to lie in the table for a sufficient timeout before making space—fast entry and slow outflow, making it hard to keep the ledger full.

later, I saw real-time numbers on a soft router when the protection was fully activated: connection tracking entries were around 15,000, with UDP accounting for 70%. And this amount is exactly what my 256MB memory AX3000T can't handle, so I upgraded several times over. By the way, this issue is two different dimensions from "how many devices a router can carry"—the number of devices you can carry refers to how many devices can be attached to the wireless side, the number of sessions refers to how many connections the forwarding side can record simultaneously, and how to slim down 62 devices that article is about the former, but today it's the latter. Both exceed the limit, and the symptoms are different.

three features: first confirm whether this is the problem

The fault when the connection count is full is very distinctive but often mixes with regular network disconnections or WiFi lag. I summarized it into three characteristics, which basically boil down to it:

pay attention to distinguishing from the other two situations: if FamilyMart is completely disconnected and WeChat is no longer available, and you look for the optical modem and broadband side, the triage sequence in the of the sudden network disconnection is more accurate; If only one device is stuck and everything else is fine, that's a wireless issue, not the number of connections. There's another bonus to help you prove it: shut down the most suspicious device in the house—I turned off the NAS downloader, and after about ten minutes, the whole family recovered on its own. Once the source is pinched, entries in the table are gradually released after timeout, and can survive without restarting. This is the unique 'self-healing' of the Connection Count Meter.

want to see how much space is left in your own ledger, the backend of the two-channel

brand router basically doesn't let you see session counts. It's a bit of a pity, but it's not completely helpless.

the first route, the router is either flashed with OpenWrt or is a soft router; just look at the numbers. SSH login and type two lines:

cat /proc/sys/net/netfilter/nf_conntrack_count

cat /proc/sys/net/netfilter/nf_conntrack_ max

The first line is the current number of entries, the second line is the upper limit. If the former is next to the latter, it means it's almost full. For low-memory devices, the default limit may be only 4096 or 16384, while for large-memory machines, 65536 is common. In the OpenWrt web management interface, the "Status - Real-Time Curve" section also has an active connection count curve. Hang it for half an hour and it's clear who is causing trouble at a glance. If you see the line "nf_conntrack: table full, dropping packet" in the system log, it means the router admitted to losing packets—solid proof.

second option: branded phones have nowhere to look at numbers, so use the source comparison method. Disconnect or close the background apps one by one you suspect, and after each off, wait ten minutes to see if the whole family recovers. Once restored, it's the device. That night, that's exactly how I locked onto my NAS. On Windows computers, you can also use netstat-ano to count the current local connection count as a reference—a single computer usually has about one or two hundred connections, but if a machine itself has a thousand or two hundred connections, it's probably a software app causing trouble.

four tricks: first manage the big players, then adjust the configuration

confirm the connection count is full, and follow this order, from fixing the root cause to switching devices:

first trick is to control downloaded software. In qBittorrent, under "Tools - Options - Connections," the global maximum connection count is around 300, each seed has a maximum of 40 to 80 connections, and four to eight upload slots are sufficient. Breeders might feel sorry for the upload volume, but in actual tests, the list was limited to 300 pairs of over 200 seeds like mine, and uploads barely dropped. The router was revived first, so this bill is worth it no matter how you calculate it. I wrote the full background of this setup in the of the NAS downloader.

second move: identify the invisible big players. The prime suspect is the "P2P acceleration" switch in the TV box and video app—it uses your broadband upload to help the platform distribute videos, essentially a BT machine, with connection counts skyrocketing. My cousin's house found this one. The video app on the box had P2P acceleration enabled, and after turning it off, their router never went out late at night. Camera cloud storage uploads and cloud storage synchronization are similar. If you have many devices, you can compare device composition in the smart home article check them one by one.

third trick: Soft router users can directly expand the table. With 8GB of memory, the N100 can be limited to 260,000 entries, which only takes up about 70 to 80 MB more memory, so there's no pressure at all. The command is sysctl -w net.netfilter.nf_conntrack_max=262144. If you want to restart without losing it, write it to /etc/sysctl.conf. UDP timeout can also be reduced from the default 30 seconds to 15 seconds, and short sessions like DHT are no longer permanent. Don't mess with small memory machines—the watch itself consumes memory. A 256MB hard expansion will only squeeze memory, so it's better to just look at the next move.

fourth trick: change the machine. This is no joke—if your home has strict connection needs like conservation plants, multiple cameras, and P2P acceleration that can't be turned off, a budget phone just won't handle it. This isn't something you can fix with settings. When selecting a model, keep two things in mind: starting at 256MB of memory, and prioritize those specs that dare to specify session counts. Why do large memory require large meters? How do CPU and memory divide the tasks of forwarding? CPU and DHCP fit together perfectly.

Wrapping up, here's the troubleshooting order: symptom three, feature checkmark→ source comparison, lock the major user, → download the software, connect the number of connections→ turn off P2P acceleration→ soft router, expand the table→ If it really doesn't work, switch to a machine with larger memory. Don't be like my initial reaction—almost called for repair and asked the technician to come over. They tested the light deterioration and everything was normal, so they made a trip for nothing and owed a favor.