nfs-backed mmap file results in 1000s of WRITEs per second in RHEL 6

Solution Unverified - Updated -

Issue

  • If two (or more) processes are doing nothing more than writing to the memory addresses of an mmapped shared file on an NFS mounted file system, it results in the kernel scribbling WRITEs to the server as fast as it can (1000s per second) even while no syscalls are going on.

  • Bad Performance when many processes write on NFS file

  • Bad Performance on writing with several processes to a nfs located file
  • Writting on a mmaped files using multiple processes does not work well with Rhel6 nfs client
  • An application works with expected write performance when using RHEL5 NFS client and and performance decreased on RHEL6 NFS client .
  • We see a huge performance regression between RHEL 5 and RHEL 6 when it comes to mmap() and msync() calls to files on NFS shares.A test program that mmaps files living on NFS, writes to them and msyncs them back to NFS works up to 9 times slower on RHEL6 compared to RHEL5.
  • Performance regression with mmap on files on NFS

Environment

  • Red Hat Enterprise Linux (RHEL) 6 used as NFS client

Subscriber exclusive content

A Red Hat subscription provides unlimited access to our knowledgebase of over 48,000 articles and solutions.

Current Customers and Partners

Log in for full access

Log In