Network access constantly dying

[jbo5112]

I installed ESXi on a machine with a Realtek 8111/8168 NIC, and found some drivers to get it working. Sometimes the network connection works fine for hours and hours. Sometimes it goes down every few minutes, and I have to unplug the network cable and plug it in again to get networking operational again for just another minute or two. All the lights stay on like it's connected, but the machine acts like it's disconnected. I've tried network test from the server to the LAN and tests from the LAN to the server. Both will fail when it's down, and it doesn't come back up on its own.

I know all the hardware works acceptably. I booted to a Linux live CD and have been running a flood ping (outputs packets as fast as they come back or one hundred times per second, whichever is more) for hours, with only a few dropped packets. However, just prior to this, the network connection in ESXi wouldn't stay up for any longer than a minute. I know the network card isn't supported. Is there anything this could be, other than the drivers? Would the occasional dropped packet cause ESXi to shut down the network connection?

[DSTAVERT]

As you said it isn't a supported device and you are using a driver that wasn't supplied by VMware. Get a real supported card. The time wasted trying to get a non supported device running let alone stable just isn't worth it. You can find inexpensive server hardware on places like ebay, craigslist.

[jbo5112]

Yeah, but the downtime waiting for the hardware will be a major pain.

I have other drivers for the NIC that I can try. I'm wondering if it is going to sleep or something.

[DSTAVERT]

Realtec is a VERY low end piece of hardware. It is known to fall over under load. 100 pings per second is not a load.

[Scissor]

I installed ESXi on a machine with a Realtek 8111/8168 NIC, and found some drivers to get it working.

Not sure where you found your drivers... but perhaps there are newer drivers available that work better?

As for dropped network connections, could the problem be with the configuration of your ethernet switch? Perhaps your switch Administrator has incorrectly configured the switch port? Anything interesting showing in the switch logs?

Alternatively, I suggest spending the $35 for one of variations of the Intel Pro/1000 NIC cards. Actually, if I were you I would just buy the Intel NIC card.

[jbo5112]

This is a development machine with only 3 people or so accessing it and it's only connected to a 100Mbit switch. I wasn't trying to do a load test either, just make sure the connection stays active, but it holds up mostly well issuing 1100 flood pings to my Windows XP box on a different 100Mbit switch. I'm not leaving much headroom for the other machines that get routed through the link between switches, so without a better switch, I don't expect more load can be put on the card. I just assumed modern gigabit NIC's were all good enough for any load, but I guess I was naive.

I'm guessing there are no configuration issues that would cause my NIC to keep dying.

[jbo5112]

It's a 5-port 100mbit dumb switch in our office, so no logs or settings. I've tried two different ports, and there haven't been other problems on the switch. Both ports work fine with a Linux live CD. The networking even ran fine for the first few days of ESXi being installed.

I got the drivers from the forums at vm-help.com. There were some other drivers I could try, but I don't they include the necessary drivers for my JMicron PATA device and they were questionably large for Linux drivers. I'll see what I can dig up again. I'm glad it's somewhat easy to swap out the drivers.

[jbo5112]

I found a different driver that has been working for a few minutes. I'll update sometime later with how it's holding up. If this cheap NIC is working, I'd rather the money go to a gigabit switch.

I saw this comment for the NIC, with no further details on what to do: "Tx option for 'TCP Checksum Offload (IPv4)' in NIC advanced settings tab must be disabled." I haven't been able to find any such settings. Hopefully it's a problem with just his hardware and I won't need them.

[Rumple]

just as an fyi...if you are only using dumb switches, then its only $50 for an 8 port gbit switch and $35 for a 5 port....

[jbo5112]

I submitted this post yesterday, but somehow it got lost somewhere, probably something to do with the massive Time Warner outage.

The network connection failed after 5 hours. I tried some updated drivers, to no avail. I have an Intel NIC (about $36 with 2 day shipping from buy.com) on order and it will arrive tomorrow. Newegg is tempting me with a $33.98 (with shipping) 8-port gigabit switch from Asus, but any purchases that can wait will wait. There's not really a need at this point.

[DSTAVERT]

I wouldn't count on Intel desktop hardware being supported either. I'm not saying it won't work but . . .

[jbo5112]

It's the "Intel Gigabit CT Desktop Adapter" and VMware lists it as being compatible with ESXi 4.1 using the device driver e1000e version 0.4.1.7.1. I checked the compatibility guide before purchasing (the forums can't parse the URL for the results). As limited as their hardware support is I'm a little surprised it was supported, and half expected for them to require a VT enabled NIC.

[DSTAVERT]

Good deal. Lets hope it solves the problem.