Tests needed
From Linux NFS
(Difference between revisions)
DelerCnaze (Talk | contribs) m (rollichitrc4) |
|||
Line 1: | Line 1: | ||
+ | caerborelbo | ||
== NFS Parameter Tuning == | == NFS Parameter Tuning == | ||
Latest revision as of 12:29, 4 January 2009
caerborelbo
NFS Parameter Tuning
Create a test script that iterate through various NFS parameters and run test suites on them.
In /etc/fstab, try various values for each of the following parameters, for both NFSv3 and NFSv4:
default advanced test rsize=n 1024 performance at 512, 1024, 2048, 4091, 8192, 16384, 32768 wsize=n 1024 performance at 512, 1024, 2048, 4091, 8192, 16384, 32768 timeo=n 7 in busy network, performance at 3, 7, 15, 30, 60 retrans=n 3 in faulty network, robustness at 1, 2, 3, 6, 12, 24 acregmin=n 3 on highly loaded server, overall performance at 1, 2, 3, 6, 12, 24 acregmax=n 60 on highly loaded server, robustness and overall performance at 15, 30, 60, 120, 240 acdirmin=n 30 on highly loaded server, overall performance at 5, 10, 15, 30, 60, 120, 240 acdirmax=n 60 on highly loaded server, robustness and overall performance at 15, 30, 60, 120, 240 actimeo=n - ?? retry=n 10000 try with value of 1 to verify it still works correctly namlen=n 255 ?? not sure if this is worth testing port=n 2049 set server to use port 10000, and verify this options works mountport=n set mountd to use port 10001, and verify this option works mounthost=name ?? mountprog=n 100005 ?? mountvers=n 1 ?? nfsprog=n 100003 ?? nfsvers=n 2 try with 3 and 4, and verify the right version gets used nolock off ?? bg off switch this on, and verify mounts occur properly in background fg on soft off switch this on, and verify an I/O error is generated when mounting a nonexistant server hard on intr off test how this affects operation when major timeouts occur with hard mounts posix off turn on and test with filenames > 255 chars nocto off test how this works when noac off test performance with 1-N clients writing to the same filesystem on the server tcp off verify this option is ignored in nfsv4 udp on verify this option is ignored in nfsv4
Mount Parameter Testing
- Measure total time when using mount -F vs mount for a lot of mount points
- Verify that mount -vf does not actually mount the filesystem but just pretends to
- Make /etc read only, and verify that mount -n allows mounting without writing in /etc/mtab
- Verify that the -s option allows invalid -o options to be specified to mount, and be ignored properly
- Verify that mount -r mounts read only
- Verify that mount -a -O no_nfs4 mounts everything but the nfs4 mounts
- Verify that mount --bind works to remount a subtree somewhere else
- Verify that mount --move works to move a subtree someplace else
- Verify that mount exits with the proper codes on error:
0 success 1 incorrect invocation or permissions 2 system error (out of memory, cannot fork, no more loop devices) 4 internal mount bug or missing nfs support in mount 8 user interrupt 16 problems writing or locking /etc/mtab 32 mount failure 64 some mount succeeded
Umount Parameter Testing
- Verify umount -n does not write to /etc/mtab
- Verify umount -r remounts read-only when unmounting fails
- Verify the umount command notifies the server when an NFS filesystem is unmounted. (Bug in NFSv3)
- Verify that umount -t nfs4 unmounts a mountpoint only if it is an nfs4 mountpoint
- Verify that umount -t no_nfs4 does not unmount a mountpoint if it is an nfs4 mountpoint
- Verify that umount -a -t nfs4 unmounts all nfs4 mounts
- Verify that umount -a -t no_nfs4 unmounts everything except nfs4 mounts
- Verify that umount -O works to only allow mount points with the given options are unmounted
- Verify that umount -f works to force unmount of unreachable NFS systems
- Verify that umount -l detaches the filesystem from the filesystem hierarchy immediately, and cleans up references to the filesystem as the processes using them complete