I see that project tracking isn't setup yet.

I've requested it from the project admin.

Also for testing, I recommend setting up a small VM with Apache and using an external machine (Not on the same HyperVisor Server) to run AB (apacheBench) to record/stratify the statistics.

It will record request/sec, request time length, shortest, longest, failed, etc...

You'll also be able to record request concurrency.

These things my team will try to put together hopefully in the coming weekend.

AB is the closest thing you can get to a 'real world' scenario. I don't recommend using dd to write out to a network drive.

I have updated the r8168 driver:

http://www.icjacobsen.com/dld/oem.tgz

it is now a backport to ESXi of the latest r8168.011.00 driver and on top of that has extra protection built in to avoid package dropping.

For anyone who has observed problems with the r8168 driver, please test it.

Regards
Jens

For me new driver didn't help. The same situation as with old one - connection to vncserver on Solaris 10 64bit VM doesn't work, SCP transfer to Solaris 10 64bit VM reset, download from internet from Solaris 10 64bit VM doesn't work. However i can ping this VM and also ssh to it w/o any issues.

I don't know if this might help you but in trying to diagnose a timeout issue, causing my VI client to disconnect every once in a while, I stumbled upon the following.

I looked at my network link in the ESXi host configuration tab. It showed that the Vswitches were setup right but the physical NIC (Intel Pro/1000 GT) had an red X next to it. Even though all the VM's are able to communicate out it and, of course, I could get to it with VI, WinSCP and Putty. When I pulled up the properties, I hard set the connection speed from Auto Negotiate to 1000 Full Duplex.

Magically the red X went away. Things seem to be a little zippier, but I still get disconnected so, unfortunately, that wasn't 'the' problem but it looks like it was 'a' problem.

Might want to see if the physical NIC is showing any health issues in the VI.

George

Hi Jens,

Thank you very much for making this driver for the Realtek card available to the VMWare community, I for one really appreciate it!

I've installed ESXi 3.5 U4 on a Gigabyte EX58-UD4P motherboard which has the Realtek 8111D Gigabit NIC. Your driver seems to work well for connecting to ESXi and administering itself (SSH, VM Infrastructure Client, etc), however I also seem to be experiencing slow and flaky network performance when dealing with the VM's themselves.

The installation of Ubuntu x86 9.04 onto a VM using the Infrastructure Client from an ISO located on my local workstation bombed out just after half way through with an error that indicated installation media error. Also have issues with a Windows 2003 Server R2 x86 install which won't allow me to activate the install of windows. I also get 400 errors when trying to browse the Internet using IE6, however I am able to obtain DHCP leases, ping, trace route or do nslookups without any issue from the VM's. I tried setting the network speed down to 10Mb, half duplex just in case it was a cabling issue, but no luck.

Apart from using your driver and setting the network speed to 10Mb half duplex, my install of ESXi is using default values. Any ideas?

Does the speed/duplex match what is set on the physical switch that you're connecting to? Do your VMs have any trouble with the network if the traffic stays on the LAN? Any VLANs in use?

Hi Dave,

My switch is just your basic 10/100 variety which usually connects at 100Mb full duplex. I've actually tried connecting using all link speeds, both at the ESXi and VM level with no difference.

VM's have less trouble if the traffic stays on the LAN using the one class C subnet, however the performance is still extremely slow and have a lot of issues with what seems to be corrupted data going on the error message given when trying to install Ubuntu from an ISO over the network. I am also unable to connect to a VM directly using VM Infrastructure Client (using name or IP), however if I connect to the management network first and then right click on the VM and select "open console" I am able to connect. Looking at the output from "netstat -ano" at the command line I can see there are only connections to the management network IP address from the VpxClient.exe process, so I'm guessing that connecting to a VM directly is different to connecting to a VM via the management console?

Apart from very basic network function (ping, trace route, etc) nothing seems to work when trying to communicate with hosts outside the LAN. Originally I didn't configure any VLANS, but I thought I'd give it a try last night and now the NIC's in the VM's won't connect at all (showing media disconnected in Windows 2003 VM). This could well be a config issue on my side of things though, however I have set everything back to original settings now and the problem persists.

I think the driver was compiled for the 8111c NIC first so it may had a problem with the 8111d. Would you have a 3Com 905, Intel Pro100/100, Broadcom 57xx around to try?

I ended up getting an Intel Pro 1000GT, and everything is working perfect now. I've left the 8111D as a standby adapter in case the Intel fails, and hopefully I'll have more luck with it as the driver matures.

Bridged networking and dropped packets are fixed!!! The updated driver ended up fixing our sluggish performance (at least in 32-bit OS VMs - haven't tested on 64-bit). Just to clarify, the problem of sluggishness started to occur when we installed VMware Tools on the vm. Otherwise, the VM performed without network sluggishness (despite the slow network adapter). With vmware tools installed, we couldn't even reach the internet via http requests.

We just started testing this today, and hopefully the problem is permanently fixed. If there are any issues, I will report them back in this forum posting. Hopefully, you won't hear from me!

Thanks so much, Jens. Your efforts are much appreciated.

Andy Slezak