NFS utils package, and an fsx question

DISCUSSION Tom McNeal 2 participants 3 messages May 13, 2002

Subsystems (2)

NFS client NFS server

lore

Share & tools

lore

Tom McNeal

author

2002-05-13 02:25 UTC #0 lore raw

Hi -

I've created and tested a new NFS utils package, which can be downloaded from

http://nfs.sourceforge.net/nfs-utils-1-0-1-pre7.tar.gz

The tests consist of the 4 2002 Connectathon tests suites (basic, general, special, and lock), fsx, a duplicate request cache exerciser, and bonnie++, all running simultaneously overnight. I tested a 2.4.18 and 2.2.19 UP client against 2.4 and 2.2 UP servers. The failure rate was less than 1%, primarily due to interaction between special and general tests that I haven't quite figured out yet.

I see nothing in the utils package that tells me it isn't ready for prime time. I'd really like to see someone with MP platforms pick this up and test it, but other than that, I think its ready to rename as release 1.0.1. BTW, I don't know the history of release 1.0.0; It was already 1.0.1-prex when I started working on it.

I have a couple of questions about fsx behavior:

- It appears that with memory mapping allowed, fsx doesn't
allow more than 12 operations running per test, regardless of the kernel; is this fixed with a later version of fsx, or maybe I just have my system misconfigured? My version of fsx is about 6 months old, and I didn't look at configuration at all.

- In 2.2 kernels, the failure rate climbs significantly
unless I cut the number of operations per test in half; once I turned off memory mapping, I ran with 5000 ops in the 2.4 kernel, but had to run about 2500 ops in the 2.2 client kernel to get the same failure rate (~1%). Is this a suprise?

Tom

--
------------------------------------------------------------
Tom McNeal [email protected] (650)906-0761 (cell)
------------------------------------------------------------

_______________________________________________________________

Have big pipes? SourceForge.net is looking for download mirrors. We supply the hardware. You get the recognition. Email Us: [email protected]
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs

Trond Myklebust

2002-05-13 05:35 UTC #1 lore raw

Quoted 1 line

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
" " == Tom McNeal <[email protected]> writes:

> The tests consist of the 4 2002 Connectathon tests suites > (basic, general, special, and lock), fsx, a duplicate request > cache exerciser, and bonnie++, all running simultaneously > overnight. I tested a 2.4.18 and 2.2.19 UP client against > 2.4 and 2.2 UP servers. The failure rate was less than 1%, > primarily due to interaction between special and general tests > that I haven't quite figured out yet.

Any details on the failures? 2.4.18 was tested extensively at Cthon with no problems being observed...

Cheers,
Trond

_______________________________________________________________

Tom McNeal

author

2002-05-13 15:21 UTC #2 lore raw

What I see is this:

- The tests fail most often when they are the only tests
running, and nothing else is going on. Generally, when I'm doing the test run, I have all tests running 500 times, and I might see 2 or 3 of these failures.

- The special tests fail with this error:

--
SPECIAL TESTS: directory /mnt/first/usr/local/NFSTest
mkdir: cannot create directory `/mnt/first/usr/local/NFSTest': File exists cd /mnt/first/usr/local/NFSTest; rm -f runtests runtests.wrk READWIN.txt Makefile op_unlk op_ren op_chmod dupreq excltest negseek rename holey truncate nfsidem nstat stat stat2 touchn fstat rewind telldir bigfile bigfile2 freesp cp runtests runtests.wrk READWIN.txt Makefile op_unlk op_ren op_chmod dupreq excltest negseek rename holey truncate nfsidem nstat stat stat2 touchn fstat rewind telldir bigfile bigfile2 freesp /mnt/first/usr/local/NFSTest runtests.wrk: runtests.wrk: No such file or directory
special tests failed
--

- The general tests fail with this error:

--
GENERAL TESTS: directory /mnt/first/usr/local/NFSTest
mkdir: cannot create directory `/mnt/first/usr/local/NFSTest': File exists cd /mnt/first/usr/local/NFSTest; rm -f Makefile runtests runtests.wrk *.sh *.c mkdummy rmdummy nroff.in makefile.tst
cp Makefile runtests runtests.wrk *.sh *.c mkdummy rmdummy nroff.in makefile.tst /mnt/first/usr/local/NFSTest

Small Compile
runtests.wrk: line 85: unexpected EOF while looking for matching `"' runtests.wrk: line 88: syntax error: unexpected end of file general tests failed
--

I think the issue is distructive interaction between the two tests, since I am running them in tandem, rather than in succession, as is done normally. They generally fail together, either once or twice each, and then start completing successfully again.

Tom

Quoted 11 lines · Trond Myklebust

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
" " == Tom McNeal <[email protected]> writes:
> The tests consist of the 4 2002 Connectathon tests suites > (basic, general, special, and lock), fsx, a duplicate request > cache exerciser, and bonnie++, all running simultaneously > overnight. I tested a 2.4.18 and 2.2.19 UP client against > 2.4 and 2.2 UP servers. The failure rate was less than 1%, > primarily due to interaction between special and general tests > that I haven't quite figured out yet.

Quoted 5 lines

Any details on the failures? 2.4.18 was tested extensively at Cthon with no problems being observed...
Cheers,
Trond

--
------------------------------------------------------------
Tom McNeal [email protected] (650)906-0761 (cell)
------------------------------------------------------------

_______________________________________________________________