Server Fault is a question and answer site for system and network administrators. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

File operations are necessarily going to be slower over NFS. What I would like are some rough estimates of expected slowdown between local and NFS file operations, specifically for file creation when working on a properly tuned NFS file system.

I am noticing our NFS file system is 1,000 times slower for file creation (therefore nearly unusable when copying over large directories, for example). As I help our system administrators with this problem, I would like some expectations to serve as a baseline as we measure progress.

Adding additional info per Timothy's request:

NFS version 4 Protocol

ping stats from client to NFS server:

8 packets transmitted, 8 received, 0% packet loss, time 6999ms
rtt min/avg/max/mdev = 0.167/0.216/0.244/0.029 ms

io (input/output): 147584499 3271482498

The NFS file system is consistently 1,000 times slower. It is not simply a function of what is happening on the network.

Post Discussion Update

We have discovered a NFS configuration issue. By changing the exports to async from the default of sync, we have gained 60x increase in performance. async does come with a risk as the client does not wait for the data to be written to disk before preceding, so some files may become corrupted but the risk is minimal.

share|improve this question
1000 times slower seems high, but there are so many factors. A few major ones to consider are: network latency between client and server, NFS protocol version, I/O size, and how busy the NFS server is. Including that information, or correlating it with your measured performance, will help better with understanding and predicting performance. – Tim B Mar 6 '13 at 21:41
@TimothyButler thanks for your help. I try to provide additional relevant information above. – Julien Chastang Mar 6 '13 at 22:25
Sub millisecond RTT means it's not latency. 1000X slower can refer to latency or bandwidth. Expected slowdown itself is hard to state. For example, I'd expect writes to a fast NFS filer with large write cache to be faster than writes to a slow local disk, such as a low RPM green or laptop disk. Or, such as a test I just did, NFS could be slower. I just got 55MB/s off a VM, through a firewall to a filer. Local to RAID in the ESX server was 350MB/s. Capture NFS with wireshark to see if there is anything obvious. (I'd also try NFS v3, just to see if it's a v4 protocol/config issue) – Tim B Mar 7 '13 at 15:02
@TimothyButler as you can tell I am a novice in this area so this discussion is helpful. So you have a factor of ~7 slowdown over NFS. That is the rough estimate I am looking for. Thanks again. – Julien Chastang Mar 7 '13 at 15:38

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Browse other questions tagged or ask your own question.