Tests performed
NFSv4 is stressed using fsstress in various configurations, exactly in same conditions than with kernel 2.6.11rc4, but we are now using the final 2.6.11 kernel with CITI patches.
Test one : fstress:
fsstress is run in various configurations and parameters
Tests are done with various number of subprocess : 1, 2, 4, 8, 16, 32
Tests are done with a various lenth of instructions lists: 1, 2, 4, 8, 16, 32, 64, 128
Test two: (thanks to Dumbledore)
Run dd if=/dev/zero of=/mnt/nfs/fichier_test one hour
Conclusion
Main bugs from 2.6.11rc4 are fixed.
Only a few bugs are remaining
While runing fsstress with more than 10 process, a 1000 operations length list during 20 minutes
- functions open_path and path_create (that match to a open) are going slower and slower
- After a long while, tests are blocked.
- On the server we see a symlink with 8 caracters name pointing to a 2732 caracter path that does not exist
-
Conclusion : probable bug on path resolution.
Runing the second test 30 minute. The file fichier_test is growing up to 32GB. We can not kill and umount properly the nfs file system.(umount is blocked)
Client message : kernel: VFS: Busy inodes after unmount. Self-destruct in 5 seconds. Have a nice day...
Software configuration
- Server :
- Export options are :
/export
<world>(rw,wdelay,insecure,no_root_squash,no_subtree_check,fsid=0)
- mount options, NFSv4 :
rsize=32768,wsize=32768,timeo=14,intr,noauto
- fsstress -l 0 -d /mnt/benchs_nfs/tmp/ -n <operation_sequence_lenght> -p
<number_of_process> -r
- Linux 2.6.11-CITI_NFS4_ALL-1
hardware configuration
Client
- 2 processors : Intel(R)
- Xeon(TM) CPU 2.80GHz, cache 512 KB
- Total memory: 2Gb
- Ethernet: 1Gb/s link
- Distribution : modified Fedora Core 2
Server
- 4 processors : Intel(R) Xeon(TM) CPU 2.80GHz, cache 512 KB
- Total memory: 24Gb
- Ethernet: 1Gb/s link
- Distribution : modified Fedora Core 2
