Client for NFS using increasing amount of CPU

Question

We have an issue with Microsoft’s client for NFS sometimes using a large amount of CPU (50-100%) when accessing a NFS share on a linux box.

The server seems to get in a state where accessing the NFS starts to use an increasing amount of CPU to do the same operation, causing the system to slow down. Once it gets into this state it does not seem to ever recover. A reboot of the system does fix it temporarily. Any guidance would be appreciated about methods to diagnose, possible causes, and even better possible solutions.

Setup:

We have several win 2008 R2 cloud servers accessing an NFS share (redhat cluster which access a NFS). They all use client for NFS to access it, using the registry fix described here: http://blog.cuongnv.com/2009/11/windows-7-client-for-nfs-and-user-name.html, we do not have either user name mapping or active directory. They access it via an IP rather than a mounted drive, which in my understanding still uses Microsoft client for NFS to access the NFS share on the linux box. Each server has the drive mounted so we can see what files are on the NFS.

For reference purposes we also have two linux cloud servers accessing the NFS, but have not noticed an issue with them.

Usage: All files about 8-15KB The NFS is getting a file listing of several directories containing a total of about 8500 files every minute and then individually checking the date of the file. About 26 times a minute a subset of this directory with about 640 files is listed and each file checked for date. Also there are about 60 file writes and 60 file deletes every minute.

As a side note we are also experiencing (A probably unrelated) issue where the servers have a stop error when accessing the NFS. We believe this will be fixed by this patch: http://support.microsoft.com/kb/2584874 (not yet applied).

Answer 1

Sorry for the bump, but finally have a semi-solution.

This was to increase the NFS threads on the server and improve the efficiency of the code so it is not doing as many directory listings. These combined do allow it to work reasonably (at least in our situation).