We have found at more than 7 sites whom we have built on or upgraded to ESXi 6 and we are using the VMXNET3 adapter extreme slowness to SQL and other database applications. Here is what we have gathered so far.
ESXi 5.5u2 or newer including all version of 6
Windows 8 or newer / Windows 2012 or newer
VMXNET3 adapter on client side using one of the above OS's
SQL or other Database application on Server
SQL or other Database application and Client are on different hosts (condition does not exist on same host and does not affect physical clients accessing the database)
Change to E1000 adapter and the problem goes away
We are able to reproduce the issue at more that 7 client sites.
Anyone else seen this behavior and have a fix that does not include removing the VMXNET3 adapter?
Did you have a chance to review the below KB:
Also, if you notice packet loss on Guest level, you can increase ring buffer size:
I have recently had the exact same thing happen to us (Server is a Windows 2012 R2 server, RAID controller is an LSI MegaRAID SAS 9266-4i). I tried everything in the book I could think of (different NICs, differnt hard drives, changed to SSD drives. In each case nothing worked. All databases accessed over the network were as slow as a dead dog.
I also tried all the solution SavkoorSuhas mentions below (VMware KB: Poor network performance or high network latency on Windows virtual machines). I did not try the other solution he mentions ("Also, if you notice packet loss on Guest level, you can increase ring buffer size:VMware KB: Large packet loss at the guest OS level on the VMXNET3 vNIC in ESXi").
When I changed to the E1000E adapter, everything is back to normal.
Hi Sean, did you have any updates regarding your investigation? We're seeing this issue too between front-end servers and their SQL back-ends. The issue goes away if we put the VMs on the same host, but returns as soon as we separate them. Switching to E1000 also negates the issue, but obviously not the fix.
Just curious where you'd left off on this.
I know this is an old post, but was wondering if anyone is still having issues with the latest version of 6.0? We are running 6.0, 5224934 and are seeing some similar issues. We have not yet tried rolling back to an E1000 adapter as it seems to only be an issue on one SQL VM.
Yes, we have the latest version of VMware tools installed. We do have other SQL servers that are virtualized and this is the only one we are seeing issues like this. From a VMware side everything looks great. Very low latency, no huge CPU spikes, Lots of memory and it is not overcommitted.