Re: strace on smb/iozone

Bill Hawes (whawes@star.net)
Wed, 12 Nov 1997 10:34:17 -0500

Steven N. Hirsch wrote:
> It gets better all the time :-(.
>
> open("iozone.tmp", O_RDONLY) = 4
> write(1, "Writing the 1 Megabyte file, \'i"..., 62) = 62
> times({tms_utime=2, tms_stime=43, tms_cutime=0, tms_cstime=0}) = 12162725
> read(4, "", 512) = 0
> ^^^^^^^^
>
> write(1, "Reading the file...Error reading"..., 41) = 41
> _exit(1) = ?
>
> For whatever reason, read() returns a zero length on the first call.
> Yet the file does exist. This suggests that something in the VFS or
> smbfs layer is in a race with the call and has not updated its notion
> of the file's state. FWIW, Bonnie behaves the same way; failing on the
> first read attempt.

Hi Steve,

I've compiled and run the copy of iozone you sent me, but haven't been
able to replicate the read error problem.

I get write speeds scaling from 350kb/sec for 512 byte blocks up to 2.8
Mb/sec for 8K blocks. At the low end, the difference from 512 to 1024
byte blocks is nearly linear, so the slow operation appears to be
dominated by network overhead. (No surprise there.)

Read speed goes from 29Mb/sec to 52Mb/sec over the auto test series, so
the page cacheing seems to be working well.

I'll continue looking for the Win 3.1 read error problem, but am quite
puzzled that the smbfs debugging trace didn't show any attempt to read
the file. Any chance it could be a permissions problem, or something
like that? Can other utilities read the file iozone creates, and do
ordinary commands (cat, vi, etc.) work with the Win 3.1 share?

Regards,
Bill