BUG: acme_fcase_trace negative counter values

[nawtrey]

It looks like the acme_fcase_trace log has some negative data for the POSIX_F_META_TIME counter. I noticed this while attempting to write a regression test for issue #590.

Here is some output illustrating the problem:

module:  POSIX
Negative values:  [-3.06159650e+171 -1.72714531e-307 -8.63423493e-307 -4.79982338e-306]
by-average data: 
Read      1.624701e-01
Write     4.472180e-02
Meta    -3.737300e+167
----------------------
module:  MPIIO
Negative values:  []
by-average data: 
Read     0.157147
Write    0.436347
Meta     0.003968
----------------------
module:  STDIO
Negative values:  []
by-average data: 
Read     0.067240
Write    0.000083
Meta     0.042183
----------------------
Final I/O cost dataframe:
           Read     Write           Meta
POSIX  0.162470  0.044722 -3.737300e+167
MPIIO  0.157147  0.436347   3.967560e-03
STDIO  0.067240  0.000083   4.218273e-02

Basically the POSIX_F_META_TIME counter has a handful of negative values, one of which is very large, and this is resulting in a large, negative sum for the Meta time. Maybe @shanedsnyder can take a quick look at the counter data to confirm, but for now I'll probably look at using a different log.

[carns]

Wow, that field is really way off:

POSIX   -1      2990992946131980000     POSIX_F_META_TIME       -3061596495762055327894679314511836560340477487190259516241843015484138233235383091501060635977155859197306952421967514327453521398186444153083051710408022781116533717336064.000000    /gpfs/mira-fs1/projects/radix-io/snyder/acme/test_F_case_cetus_dxt/run/drv_flds_in      /gpfs/mira-fs1  gpfs

The rest of the record for that drv_flds_in file doesn't look particularly remarkable, though. It was opened 8K times and read 16K times.

I'm not coming up with many ideas for how to work out what went wrong here from looking at the log.