RE: pmlogger -u questions

["Ken McDonell" <kenj@xxxxxxxxxxxxxxxx>]

> -----Original Message-----
> From: Frank Ch. Eigler [mailto:fche@xxxxxxxxxx]
> Sent: Friday, 18 April 2014 8:52 AM
> ...
> It would be informative to know how this translates to disk i/o.  We don't
use
> O_SYNC or fsync at this point, so it would be nice to be reassured that
we're
> not triggering many more physical I/Os than before.  (We'd just be pushing
> buffering to the kernel.)

These tests are not really long enough, and it is hard to isolate physical
writes cause by pmlogger from the general background noise and whatever the
kernel is using as the algorithm du jour to flush dirty blocks to disk.

And we are not using anything to trigger more aggressive or pre-emptive I/O
as you note.

However I did do one quick test and the number of disk writes writes for the
short case was 290 for the 3.9.2 version and 265 for the new version ...
given the factors outside my control, I'd be willing to declare this a
"draw" (the expected result) and certainly does provides no support for the
hypothesis that something really bad has gone wrong here.