[dune-pdelab] [dune-pdelab-commit] dune-pdelab r2050 - in trunk: . dune/pdelab/common lib m4

[Jö Fahlke]

Am Wed, 23. May 2012, 14:06:39 +0200 schrieb Christian Engwer:
> isn't this so fundamental functionality, that it should go into
> dune-common?

Yes, probably, sooner or later, but I'd like to let it mature here for a
while.  I need it in PDELab for some timing measurements right now.

There are some open questions:

 * Need a data type with sufficient precision for temporal values.  Modern
   systems provide nanosecond resolution[1], and you need to be able to
   accurately substract temporal values of, lets say, up to 60 years.  long
   double is an obvious candidate, but is it actually large enough on all
   relevant architectures?

 * clock_gettime() is an optional part of POSIX, so are there relevant systems
   that don't implement it, what to do in that case?

 * The manpage warns about the effect of process migration on the process time
   obtained with clock_gettime().  I did some tests with forced process
   migration and only observed some jitter which I would deem acceptable[2].
   But still, I was only able to test for i386/amd64 so the effect may be much
   worse for other architechtures.

Bye,
Jö.

[1] Although the actually accuracy is usually much worse.
[2] In any case much better than anything that can be obtained by getrusage().

-- 
Jorrit (Jö) Fahlke, Interdisciplinary Center for Scientific Computing,
Heidelberg University, Im Neuenheimer Feld 368, D-69120 Heidelberg
Tel: +49 6221 54 8890 Fax: +49 6221 54 8884

Der Horizont vieler Menschen ist ein Kreis mit Radius Null - und das
nennen sie ihren Standpunkt.
-- A. Einstein
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 828 bytes
Desc: Digital signature
URL: <https://lists.dune-project.org/pipermail/dune-pdelab/attachments/20120523/21f6b2f6/attachment.sig>