#506 Coarsening of ALUGrid very slow on large grids

Metadata

Property	Value
Reported by	Uli Sack (usack@math.fu-berlin.de)
Reported at	Feb 23, 2009 17:36
Type	Bug Report
Version	Git (pre2.4) [autotools]
Operating System	Linux
Last edited by	Carsten Gräser (graeser@math.fu-berlin.de)
Last edited at	Sep 15, 2009 15:28
Closed by	Carsten Gräser (graeser@math.fu-berlin.de)
Closed at	Sep 15, 2009 15:28
Closed in version	Unknown
Resolution	Fixed
Comment	Seems to be fixed. However it would be interesting what fixed this problem.

Description

Hi Dune!

I have had an ALUGrid (sequential) of roughly 18mio nodes that was coarsened in two successive marking and adapting cycles to about 4.8mio nodes. This took about 14 hours (!!!) on a 64Bit SuSe10 system running on a machine with 256GB RAM (and 8...) model name : Dual-Core AMD Opteron(tm) Processor 8220 cpu MHz : 2792.922 cache size : 1024 KB Same thing for about 9mio to 2.5mio nodes took round about 3,5h, 4.9mio->1.3mio 24min. This has a somewhat quadratic feel. Adding some stopwatches showed that more than 95% of that time goes into the preAdapt(),adapt(),postAdapt() cycle. I guess it actually goes into the adapth() method but I'm checking on that right now.

Would you happen to know where the implementation of the coarsening might cause quadratic (or let's say nonlinear) runtime?

tnx, Uli