Comprehensive data protection for all workloads
Post Reply
zoltank
Expert
Posts: 229
Liked: 41 times
Joined: Feb 18, 2011 5:01 pm
Contact:

Move away from VMXNET3?

Post by zoltank »

With the latest issue with VMXNET3, what is the NIC I should be using for my VMs? Is the e1000 the safest choice?

If I should stick with VMXNET3, is there a recommended configuration to use?
wa15
Veteran
Posts: 323
Liked: 25 times
Joined: Jan 02, 2014 4:45 pm
Contact:

Re: Move away from VMXNET3?

Post by wa15 »

To be honest with you, I am not sure how wiodespead the issue with the VMXNET3 is in production environments. We have that vmnic in 150+ VMs and so far have not had an issue. We have to remember that there are also a number of big benefits using VMXNET3 instead of the e1000 and some of the other vmnics.
chrisBrindley
Enthusiast
Posts: 43
Liked: 3 times
Joined: Aug 21, 2013 1:15 pm
Contact:

Re: Move away from VMXNET3?

Post by chrisBrindley »

I agree, there are Millions of vmware servers on vmxnet3 nic drivers, if this issue was major then vmware would have released a patch by now, I have 1500vm servers running on vmxnet3 without issue, cant say the same for E1000
Gostev
Chief Product Officer
Posts: 31460
Liked: 6648 times
Joined: Jan 01, 2006 1:01 am
Location: Baar, Switzerland
Contact:

Re: Move away from VMXNET3?

Post by Gostev »

Agree, VMware is generally very good at releasing patches.

The main point here is that over the past few years, VMXNET3 has had too many bugs comparing to other adapter types - and has been causing us much pain from technical support perspective. So, I simply don't trust it based on historical performance. May be VMware will stabilize it eventually, but with PSOD and data corruption (two different issues) this year alone, it's definitely not there for me just yet.

And yes, it may work fine on millions of VMware servers, because vast majority of VMs out there are very lightly loaded. It is those heavy production applications (often mission critical) where it typically starts to misbehave. At least the two issues mentioned above were associated with heavy network load (such as from backup activity). I guess vNICs are just not tested well by VMware in these kind of conditions... because we never had to do anything special to run into these issues in our QC labs, other than starting a few backup jobs ;)
freddy.grande
Lurker
Posts: 1
Liked: never
Joined: Mar 20, 2014 5:15 am
Full Name: Freddy Grande

Re: Move away from VMXNET3?

Post by freddy.grande »

Hmm that's interesting. What issues are these?
Sorry for being lazy but just looking for KB articles etc to assess our usage of it.

I recently had to move away from E1000 to VMXNET3 on a few servers to fix some issues.
E1000 was causing PSODs on ESXi 5.5 when copying files through Windows Explorer (fixed in 5.5U1).
MS System Center Configuration Manager was deploying packages to distribution points where the hashes didn't match. Apparently this can occasionally be caused by failing NICs and "some" virtual NICs. Changing it resolved the issue.
tom11011
Expert
Posts: 192
Liked: 9 times
Joined: Dec 01, 2010 8:40 pm
Full Name: Tom
Contact:

Re: Move away from VMXNET3?

Post by tom11011 »

I use vmxnet3 for Receive Side Scaling in windows (RSS), definitely required for a busy VM! For those not familiar with RSS, here is what it does.

On a virtual machine, there is no hardware network card. The virtual machine uses cpu for its network card. Without RSS, no matter how many cpu's your virtual machine has, the network is always tied to cpu0. If cpu0 is busy (maxed out at 100%), then network packets will drop even if nothing is going on with the other cpu's. With RSS turned on in windows, network load is spread across all the cpu's not just cpu0.

Turning on RSS is more than just installing vmxnet3 in vmware. You then have to go to the configuration section of the adapter in windows and enable it on a drop down menu. If you are using Windows 2003, there is an extra step but I don't recall what it is.
RossFawcett
Service Provider
Posts: 18
Liked: 4 times
Joined: Jul 14, 2014 8:49 am
Full Name: Ross Fawcett
Location: Perth, Western Australia
Contact:

Re: Move away from VMXNET3?

Post by RossFawcett »

I've personally had more issues with E1000's than with VMXNET3 but that may just be me getting lucky.

You mentioned PSOD's with VMXNET3 whereas we saw the opposite, we were hit with the PSOD issue and the GRE/PPTP problems at a number of sites using E1000's. All were either upgraded or migrated to ESXi 5.5. This was fixed in a hotfix though.
kjstech
Expert
Posts: 160
Liked: 16 times
Joined: Jan 17, 2014 4:12 pm
Full Name: Keith S
Contact:

Re: Move away from VMXNET3?

Post by kjstech »

No issues here on ESXi 5.0, 1489271 and 1851670.

All our Windows based VM's with e1000's have been migrated to VMXNET3, including our templates for when we spin up new VM's. I did read a few VMWare whitepapers that pointed out the benefits, so that was the inspiration for the migration. Never had an issue here.
Gostev
Chief Product Officer
Posts: 31460
Liked: 6648 times
Joined: Jan 01, 2006 1:01 am
Location: Baar, Switzerland
Contact:

Re: Move away from VMXNET3?

Post by Gostev » 1 person likes this post

Since someone above has asked, I went through my records and the following were the issues with VMXNET3 causing us support cases:

1. Interop issue with Windows Server 2008 R2 and Windows 7, which causes static IP address assignment loss upon restore and failover. This is by far the biggest, just search this forum for VMXNET3 to see all the threads this issue has caused. Worst thing is that impacts full VM restores, and can only be fixed by patching each Windows server BEFORE you create a backup.

2. VMXNET3 and Jumbo Frames on Windows. This bug can cause performance problems, instability and loss of connectivity (fixed by now).

3. This new interop issue that actually sparkled this discussion. Again, this should go away as people move Window Server 2012.

I guess my attitude to VMXNET3 mostly comes from the first issue, and endless support cases it caused in the past years... but this one should be slowly dying out as people move to Windows Server 2012 and Windows 8.

Now, apparently I was wrong about those critical PSOD and data corruption issues from earlier this year. My records show they were NOT related to VMXNET3, but rather E1000E (my apologies). So perhaps VMXNET3 indeed makes it a safer choice these days than the default E1000E.
Post Reply

Who is online

Users browsing this forum: DataDefender and 322 guests