Test
I try to analyse the causes of the NFS behaviour I saw here. The first step is to compare NFS performances with local file system (ext3) performances for file creation. The script used is the same in the both tests.
Diagrams:
- The first diagram is the time to create n files on ext3 O(n^3).
- The second diagram is the time to create n files on NFSv4
There is no obvious relation between time to create n files on ext3 and NFSv4. In particular there is no matching between limitation when number of files is 1620000 on NFSv4 and ext3.
Be aware to the scales of the diagrams. See superposed diagram below for a more clear relation between local and remote scales
diagrams:
- The first diagram is the time to create 100 new files on ext3 when the directory does content n files. We can see the clear influence of the ext3 journal/flush.
- The second diagram is the time to create 100 new files on NFSv4 when the directory does content n files. There is no matching between the ext3 and the NFSv4 behaviour. In particular, there is no special behaviour when ext3 reachs 1620000 files, and the dispersion of the data on ext3 is not changed
Be aware to the scales of the diagrams. See superposed diagram below for a more clear relation between local and remote scales
These diagrams are superpositions of above results for a more clear interpretation of the data.
Conclusion
Behaviours and limitations of NFSv4 are not caused by the local filesystem.
Software configuration
- Server :
- Export options are :
/export *(rw,wdelay,insecure,no_root_squash,no_subtree_check,fsid=0)
- mount options, NFSv4 :
rsize=32768,wsize=32768,timeo=14,intr,noauto
- Linux 2.6.12-rc4-CITI_NFS4_ALL-1
hardware configuration
(Client and Server)
- 2 processors : Intel(R)
- Xeon(TM) CPU 2.80GHz, cache 512 KB
- Total memory: 2Gb
- Ethernet: 1Gb/s link
- Distribution : modified Fedora Core 2
