Why does RHEL6.3 generate significantly more calls via NFS than SuSE 10.2 when measured by 'du' or 'dd'?
Issue
- Slower NFS performance on RHEL6.3 when comparing it to SLES 10.2
- The slowdown is being quantified by two different types of tests that represent two different types of NS operations:
- a dd of a 2GB file to the share (Which shows a WRITE of a large file):
# time dd if=/dev/zero of=/export/subdirectory/test.dd bs=4M count=250 conv=fsync oflag=direct)
- a du -sh ran against files within the share (Which shows a GETATTR of a large number of files within a directory):
# time du -sh /export/subdirectory
- With attribute caching enabled (mount option
actimeoset to non-zero value) thedutimes are in line with each other on both platforms and the dd times are about 10 seconds faster on SLES then RHEL - When attribute caching is disabled the
dutimes are about 2x longer on RHEL then on SLES and theddtimes are faster on RHEL.
Environment
- NFS Clients:
- RHEL6.3 with kernel 2.6.32-279.11.1.el6.x86_64
- SLES10.2 with kernel 2.6.16.60-0.42.5-smp
- NFS Server: NetApp filer
Subscriber exclusive content
A Red Hat subscription provides unlimited access to our knowledgebase, tools, and much more.