Poor datastore performance from Windows NFS datastore over 10Gbps

Hi Guys,

I have a setup which is running ESX 6.7.0 connected to a Windows NFS share (because…..reasons….).

The disks are enterprise SSD in an 11 disk RAID5 volume (because….reasons..). They get some pretty reasonable performance when tested alone.

The two machines are connected via Intel x540-T2 nics with CAT6 cable.

Under normal load the max I can get is around 2.5-3.0Gbps. Running iperf and tuning some windows sizes I can max the link to 10gb without doing too much.

Connecting to Windows server with Windows VM (on the same host), I can SMB at 10Gb to the same datastore. In fact it uses the same VM Host nic off the same virtual switch.

I’ve gone through things in Windows and VMWare and made sure everything is offloading, buffers are set as large as possible, all that kind of thing.

Is there anything I can do to increase performance here. Is this something anyone as seen and conquered?

I know the setup is a little weird, and we may need to reevaluate the architecture. It’s warm DR setup with some thrown together kit which works well except it could be snappier.

View Reddit by cobarbobView Source


To see the full content, share this page by clicking one of the buttons below

Related Articles


  1. I mean, you know what the problem is, right? It’s “Windows” and “NFS” in combination.

    I’m assuming you run jumbo frames on the networking.

  2. What server or system is it in?
    Are you using Windows raid or is there a real raid card?
    Why NFS over iSCSI? Both are available in windows
    How many vms are on the host with the storage server?

    2 10 gbps links is 2 GB/s or technically 2, 1 GB/s links.

    Multi pathing or link aggregation doesn’t give
    You the total sum of the band to one device. The maximum transmission is the limit of one of the physical connection. The exceptions to this are SMB 3.0 and virtualized networking. That is why in your test with the 2 vms on the same host the performance was there. If the to vms are on the same host and vswitch the traffic goes through the cpu not the next work port.

Leave a Reply