Router drops connections when packets are transmitted

I've encountered an interesting problem and it has become a very frustrating ordeal over the course of about three weeks.

A D-Link 604 router (4 ports E class, firmware version 3.20) continuously drops ethernet communications to all four ports at the slightest sign of packets from one particular workstation. However this behavior seems to be erratic. The workstation in question has functioned normally for up to a week at a time before it flips out and goes haywire again, hosing the rest of the traffic to/from the router. A few points to note...

* The ethernet connections are actually lost, meaning the windows boxes report "Network connection unplugged." for a split second and then report that the cable has been plugged back in. I can actually watch the link lights turn off and then blink back on again simultaniously.

* When it rains it pours. Once the flogging begins and the connections start dropping it takes about thirty seconds before things calm down enough to see successful transmission of TCP/IP once again. This is assuming of course that the workstation in question is either turned off or kept silent on the network without attempting to transmit any packets.

* At the start of the insanity some packets may make it through both ways but arrive VERY late in either case. This activity halts after a few seconds.

* I have tested each workstation connected to the router by itself in an attempt to narrow down the possible causes. I also tried hooking them up in pairs sending and receiving data over the network. Pairs worked fine unless the "bad" workstation was present and even then my tests sometimes succeeded because it won't always misbehave. It can take up to a week or more before the devil of a problem rears its ugly head. I wonder about this because the network load is also very erratic from week to week (these are personal computers/workstations/media players/etc).

* The "bad" workstation will operate happily by itself indefinetely.

* Often a reboot will "fix" the problem until another machine attempts to send data at the same time as the questionable machine. At this point things usually go south once again (if the problem has shown up within the past hour).

* When the problem is not acting up the network seems to operate just fine.

* The suspect computer is a Pentium III 550mhz running Windows XP Pro 2002. More specifications of this machine are available on request.

These facts lead me to believe that it may be a hardware issue (the workstation which I suspect has an on-board NIC) or the router itself may have a quarrel with that machine for whatever reason. This is just my speculation of course.

Any help would be much appreciated. I'll answer any questions you may have in an attempt to work this out. I'm at my wits end.

Thanks,


fenreer

 

I'll assume that all possible drivers (network, chipset, etc.), the OS on the suspect machine, and the firmware on the router is up to date. Also, "Network connection unplugged" is a sign of an electrical or physical level malfunction (that is, at or below the hardware level). If you weren't getting the "Network connection unplugged" error, I'd be more inclined to think that some software was causing lots of network packet collisions. This leads to the likely suspects of: faulty network interface on the suspect machine, faulty network cable, faulty network interface on router, faulty bus on home router (I'll explain the difference).

The easiest thing to check is for a faulty network cable. Make sure that the cable never bends at a sharper angle than what you'd get wrapping it around a 2-liter bottle. Make sure the insulation is intact. Make sure it doesn't get pinched by anything -- check to see if it was ever caught in a closing door or whatever. If you have a line tester, check the signal quality on the line (and make sure to jiggle the cable). Check the RJ-45 connectors on the ends of the cable to be sure the wires are correctly seated. If you don't have access to a line tester, simply replace that network cable. Alternately, exchange the one connected to the suspect machine with another cable on your LAN. If the problem moves with the cable, there's your problem. (Also, check the length of your cable. If it's more than 100m or 330 feet, it's too long. Not that this is likely.)

The next thing I'd check for is a faulty interface on the suspect machine. You say it's an on-board adapter, so while I'd normally suggest checking that the card is correctly seated or to try it in a different PCI slot, that's obviously not going to be a problem here. Test here is to put in a (preferably known functioning) NIC and seeing if the problem persists.

If neither of those work, my best guess is that there's a problem with the router itself. You can confirm this by brining in a 5th machine to use instead of the suspect machine. If the problem persists, you know it's not the suspect machine at all.

The difference between a faulty interface on the router and a faulty bus on the router (more accurately a switched router, which most home routers are) is probably trivial as far as you're concerned (since you'll probably buy a new router either way). But for the curious out there, the electronics in a switch (or any kind of Ethernet concentrator) look in most respects like you took a motherboard and filled all the PCI slots with network adapters. If just an interface fails, you can get away with not using that network adapter anymore. The rest of the switch is fine... you're just short one port. This kind of thing is taken into account on high-end concentrators. They come with four or more replaceable blocks of 8 or 10 ports that you can swap out if one goes bad (which is better than spending several thousand for a whole new device).

A damaged bus, on the other hand, is like a damaged motherboard. It doesn't matter which network adapter you use, none of them will work right.

Best way to test this is to plug the suspect machine into a different port on the router, and then not use the port that the suspect machine was connected to.

Hope this helps!

 

Oh, I also assumed there's no particular software on the suspect machine that's accessing the netowrk at the time the problems occur. The errors you're getting make it seem unlikely, but just want to be sure.