* Re: Stress testing cifs filesystem
@ 2002-10-17 19:39 Steven French
2002-10-18 2:57 ` Andi Kleen
0 siblings, 1 reply; 4+ messages in thread
From: Steven French @ 2002-10-17 19:39 UTC (permalink / raw
To: Zwane Mwaikambo; +Cc: Linux Kernel
Some observations about what to expect when fs stress testing against the
CIFS filesystem.
The NFS connectathon tests seem to work against CIFS at least until the end
when one of the final (optional) tests fails in which it tries to delete an
open file (this is not allowed with CIFS servers - don't know a way around
this yet but I have a few ideas that might work against Windows servers but
not Samba).
Windows servers do not support chmod/chown/chgrp easily using the CIFS
network protocol so the basic/test4 test has to be commented out in the NFS
test script. In theory these chmod/chown/chgrp ops could be done remotely
(sort-of) using Windows ACLs but there is more code to write for this and
it is a fairly esoteric part of the protocol which requires some more
experimentation. Also note that the cifs vfs memory mapping code is
disabled until oplock handling is more complete so I compile the nfs
connectathon tests with memory mapping disabled.
Samba does support chmod/chgrp/chown so, unlike against Windows servers,
the connectathon nfs tests run unaltered but only if the "unix extensions"
smb.conf parm is on in the server's smb.conf file and also the "delete
readonly" parm is on (deleting read-only files is tested late in the nfs
tests). Note that if you have an access mask specified on the server
(optional parms in smb.conf) it can make chmod generate less permission for
chmod than testcases might expect so best not to set an access mask on your
test shares while running the testcase.
The fsx file system stress testing also runs against the CIFS VFS to either
Windows or Samba servers (if -W -R options are specified when launching the
fsx test in order to disable memory mapping) although I am not sure that I
have ever been patient enough to run it all the way until the end. I
have tried the newer versions of LTP as well and have not found any
problems so far but have not run all the way through every test on the
current code but plan to. I would like to find a test that tests more
esoteric combinations of open flags, multiply opening the same files from
the same process as well as from multiple processes on both the same and
different machines. Not all remote filesystems pass through every open to
the remote target server (always restricting multiple opens of the same
file to a single network file open) but in file systems like the cifs vfs
that do, it would be nice to exhaustively test this useful feature. It
will be especially useful when testing oplock (distributed file caching) to
do multiple conflicting and non-conflicting opens of the same files from
both the same and different clients simulataneously.
Steve French
Senior Software Engineer
Linux Technology Center - IBM Austin
phone: 512-838-2294
email: sfrench@us.ibm.com
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: Stress testing cifs filesystem
2002-10-17 19:39 Steven French
@ 2002-10-18 2:57 ` Andi Kleen
0 siblings, 0 replies; 4+ messages in thread
From: Andi Kleen @ 2002-10-18 2:57 UTC (permalink / raw
To: Steven French; +Cc: linux-kernel
"Steven French" <sfrench@us.ibm.com> writes:
> current code but plan to. I would like to find a test that tests more
> esoteric combinations of open flags, multiply opening the same files from
> the same process as well as from multiple processes on both the same and
> different machines.
Run the LSB test suite on it. It includes most of the old POSIX/Single Unix
test suites, which test quite a lot of things and tends to find obscure
bugs in kernels and file system. It's quite complicated to setup
unfortunately. You can download it somewhere from the opengroup.org web server.
-Andi
^ permalink raw reply [flat|nested] 4+ messages in thread
* Stress testing cifs filesystem
@ 2002-10-21 16:38 Steven French
2002-10-21 20:25 ` Paul Larson
0 siblings, 1 reply; 4+ messages in thread
From: Steven French @ 2002-10-21 16:38 UTC (permalink / raw
To: linux-kernel
After struggling with setting up LSB to test
remote mounts for a while, I checked
with the LSB team on Andi's suggestion below of
using the POSIX file API section of LSB on a
network mount. They indicated that it won't
work without modifications to the LSB source,
(I had been trying to do it via just changing
the config files) something I will eventually
have to look into.
>Run the LSB test suite on it. It includes most of
>the old POSIX/Single Unix test suites, which test
>quite a lot of things and tends to find obscure
>bugs in kernels and file systemRun the LSB test suite on it. It includes
most of the old POSIX/Single Unix
Steve French
Senior Software Engineer
Linux Technology Center - IBM Austin
phone: 512-838-2294
email: sfrench@us.ibm.com
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: Stress testing cifs filesystem
2002-10-21 16:38 Stress testing cifs filesystem Steven French
@ 2002-10-21 20:25 ` Paul Larson
0 siblings, 0 replies; 4+ messages in thread
From: Paul Larson @ 2002-10-21 20:25 UTC (permalink / raw
To: Steven French; +Cc: lkml
On Mon, 2002-10-21 at 11:38, Steven French wrote:
> After struggling with setting up LSB to test
> remote mounts for a while, I checked
> with the LSB team on Andi's suggestion below of
> using the POSIX file API section of LSB on a
> network mount. They indicated that it won't
> work without modifications to the LSB source,
> (I had been trying to do it via just changing
> the config files) something I will eventually
> have to look into.
Have you tried LTP? We have several fs stress type tests in LTP and with
the (somewhat) new changes to the scripts, it's easier to specify where
the tests create their temporary directories:
'runalltests -d /mnt/cifstest'
Thanks,
Paul Larson
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2002-10-21 20:27 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2002-10-21 16:38 Stress testing cifs filesystem Steven French
2002-10-21 20:25 ` Paul Larson
-- strict thread matches above, loose matches on Subject: below --
2002-10-17 19:39 Steven French
2002-10-18 2:57 ` Andi Kleen
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.