>
>
> Ted,
>
> I pulled the latest tar file for SFS2.0 and put in on "neteng"
> in ~smq/io_benchmarks. It should have the linux patches, but I don't
> have linux clients and haven't tried compiling it. There are also
> bonnie and iobench too.
>
> The initialization code stresses parts of the system more than
> the benchmark. On the linux kernel that failed the target load point
> was only 300 ops distributed over 25 filesystems. This is a miniscule
> load by SFS standards. In theory with a single filesystem you ought
> to be able to put a load of 10 ops or so if the failure is reproducable
> with a single FS.
>
Having given up on getting the SFS2.0 client running on a 6.5.9 O2
(it currently runs on 6.4), I've managed to get the SFS2.0 stuff
running as a linux client.
It took a little work... Two of the "run/install" scripts are
missing a "`", linux "sh" caught this, plus I had to slightly
modify the #include sequence in about a half-dozen or so .c
files in the src/rpc directory, all regarding the HAVE_INT_TYPES
case, plus svc.c attempts to directly manipulate "fds_bits",
which is, by posix standards, only to be done with "FD_SET"..
So, I've started a run, will keep cranking up the knobs as far as I can.. :-)
-Ted Kline
p.s. neteng seems down, when it comes back, I'll recopy the sfs benchmark
source and produce a diff/patch for the linux differences.
|