My original nfstest script on OpenBSD was tweaked for use on Linux recently. While modifying the script, I realized that there were some shortcomings to my original script.

Rsync was the original transfer method; however, after some testing and research, I realized that this method may have a little more overhead than originally realized.

The overhead on rsync is minimal, however, I would prefer to conduct as accurate a test as theoretically possible.

While testing my FreeBSD NFS server with a Linux client, I realized that the bottleneck could very well be the server itself. The throughput I was getting was not nearly what I would expect to see. After some research, it was obvious that there were a few options available. This article is not about FreeBSD though, or I would go into more detail here. The simple answer is that I enabled vfs.zfs.cache_flush_disable="1" in /boot/loader.conf. This small change nearly quadrupled my write speeds.

I will not be posting the raw script used to generate my results within this article, however, you can view it here.


Plots

Linux NFS Interactive Plot

Linux NFS Main Effects


Has been tested on OpenBSD 6.4