OpenVZ Forum: Support » NFS Performance

Home » General » Support » NFS Performance (10x NFS Performance degredation normal?)

Show: Today's Messages :: Show Polls :: Message Navigator
E-mail to friend

NFS Performance [message #43560]

Thu, 22 September 2011 18:14

dlight
Messages: 2
Registered: September 2011
Location: Los angeles

Junior Member

First and foremost, this is my first foray into OpenVZ and I have to say I'm loving it so far. (Coming from Solaris Containers) That being said, I'm not that versed in the nuances yet.

What I have is what I believe to be a very straightforward config from a networking and resource perspective. I didn't set any VZ limits and haven't hit a single "failcnt" in user_beancounters.

I have a veth device assigned to the VZ instance and that virtual device interface is joined to a bridge on the HW node. (1Gb/Full HW link speed) Networking works just fine...

However, when conducting very basic Bandwidth tests over NFS I am getting some horrible numbers while using `cp`, but not with other utilities (dd, rsync, etc). And this only occurs when running a `cp` in a VZ instance.

Ex Data:

VZ:

NFS Mount Options:
rw,nfsvers=3,rsize=32768,wsize=32768,acregmin=0,acregmax=0,a cdirmin=0,acdirmax=0,hard,proto=tcp,timeo=600,retrans=2,addr =na1

[root@VZ /]# time cp /s55/test/1GB ./
real 1m12.103s
user 0m0.269s
sys 0m23.979s

[root@VZ /]# time rsync -vpa /s55/test/1GB ./
building file list ... done
1GB

sent 1024125088 bytes received 42 bytes 81930010.40 bytes/sec
total size is 1024000000 speedup is 1.00

real 0m11.772s
user 0m12.355s
sys 0m7.454s

HW Node:
NFS Mount Options:
rw,nfsvers=3,rsize=32768,wsize=32768,acregmin=0,acregmax=0,a cdirmin=0,acdirmax=0,hard,proto=tcp,timeo=600,retrans=2,addr =na1

[root@HWNODE /]# time cp /s55/test/1GB ./

real 0m9.543s
user 0m0.006s
sys 0m0.941s

[root@HWNODE /]# time rsync -vpa /s55/test/1GB ./
building file list ... done
1GB

sent 100884596 bytes received 42 bytes 67256425.33 bytes/sec
total size is 100872192 speedup is 1.00

real 0m1.147s
user 0m1.179s
sys 0m0.657s

I get similar results to rsync while using dd....

So what am I missing about "how" `cp` moves data, or do I need to configure a different scheduler?

All in all, I wouldn't care to much about `cp` performance, but I am seeing very similar behavior in Oracle to what I am seeing with cp. (I can't obviously switch out Oracle for rsync no matter how nice it sounds...)

Anyone have any ideas for me?

Report message to a moderator