NFS vs. CIFS

[Tabmow]

Hi All,

I'm having some interesting issues from a Linux guest with regards to transfer (mainly write) speeds using NFS and using CIFS. This is from an Oraclebox VM on a Linuxmint host which is running off the FreeNAS machine. I have also done extensive tests normally to ensure it wasn't Oraclebox and I get similar results.

Here is the output of my NFS mounts:
/media/freetab from 192.168.1.200:/mnt/tank/freetab
Flags: rw,nosuid,nodev,noexec,relatime,vers=3,rsize=65536,wsize=65536,namlen=255,hard,proto=tcp,timeo=600,retrans=2,sec=sys,mountaddr=192.168.1.200,mountvers=3,mountport=845,mountproto=udp,local_lock=none,addr=192.168.1.200

/etc/fstab:
192.168.1.200:/mnt/tank/freetab /media/freetab nfs users,rw,auto 0 0

Here is the output from the tests:

Code:

NFS: ----------------------------------------------------------------------- CrystalDiskMark 3.0.3 x64 (C) 2007-2013 hiyohiyo                            Crystal Dew World : http://crystalmark.info/ ----------------------------------------------------------------------- * MB/s = 1,000,000 byte/s [SATA/300 = 300,000,000 byte/s] 
           Sequential Read : 94.842 MB/s           Sequential Write : 13.360 MB/s          Random Read 512KB : 86.889 MB/s         Random Write 512KB : 15.320 MB/s     Random Read 4KB (QD=1) : 5.767 MB/s [ 1407.9 IOPS]    Random Write 4KB (QD=1) : 0.205 MB/s [ 50.1 IOPS] Random Read 4KB (QD=32) : 1.358 MB/s [ 331.6 IOPS]   Random Write 4KB (QD=32) : 0.224 MB/s [ 54.7 IOPS] 
  Test : 1000 MB [C: 11.0% (27.4/249.7 GB)] (x5) Date : 2014/04/19 22:31:20     OS : Windows 8.1 Enterprise [6.3 Build 9600] (x64)   

CIFS: ----------------------------------------------------------------------- CrystalDiskMark 3.0.3 x64 (C) 2007-2013 hiyohiyo                            Crystal Dew World : http://crystalmark.info/ ----------------------------------------------------------------------- * MB/s = 1,000,000 byte/s [SATA/300 = 300,000,000 byte/s] 
           Sequential Read : 81.926 MB/s           Sequential Write : 85.105 MB/s          Random Read 512KB : 73.614 MB/s         Random Write 512KB : 82.755 MB/s     Random Read 4KB (QD=1) : 6.888 MB/s [ 1681.7 IOPS]    Random Write 4KB (QD=1) : 8.690 MB/s [ 2121.7 IOPS] Random Read 4KB (QD=32) : 6.361 MB/s [ 1553.0 IOPS]   Random Write 4KB (QD=32) : 8.490 MB/s [ 2072.7 IOPS] 
  Test : 1000 MB [Z: 0.0% (0.0/5098.8 GB)] (x5) Date : 2014/04/19 22:44:26     OS : Windows 8.1 Enterprise [6.3 Build 9600] (x64)

Any ideas on where to look?

 

[eraser]

I'm not familiar with "OracleBox". Is that a different name for the VirtualBox virtualization platform? Oh, I see that the new name for VirtualBox is now "Oracle VM VirtualBox". That's a mouthful.

You mention that you are running a Linux VM, but your listed test results use "CrystalDiskMark", which is a Windows-only application.

Based upon the fact that your listed "NFS" test results are against the C: drive while the "CIFS" test results are against a Z: drive I am going to assume that you ran your tests from inside a Windows VM.

My first thought is that tests against a virtualized C: drive have additional overhead introduced by the VirtualBox virtualization stack, so it isn't really fair to compare those numbers against a direct CIFS network connection.

Ideally you should do performance testing from the VirtualBox host itself instead of inside a Guest VM. You did state that you already did extensive tests to rule out your host--can you share the results of those tests?

Are you doing anything unusual on the network side of things such as Jumbo Ethernet Frames or LACP/LAGG? We had a user a few weeks ago with bad NFS write performance that went away when he got rid of his LAGG (Link Aggregration) setup.