Am 14.01.24 um 00:36 schrieb Bill Sommerfeld via openindiana-discuss:
On 1/13/24 15:20, Andreas Wacknitz via openindiana-discuss wrote:
Am 14.01.24 um 00:13 schrieb Goetz T. Fischer:
i'm afraid making ips fast needs more than just a cython module here
and there.
don't get me wrong, that's not indiana's fault but a genernal ips
issue. maybe having a closer look at
mr. Wegmueller's go version would be a good idea after all.
If you want to make IPS fast you will probably need more than switching
the implementation language.
As Aurélien mentioned, IPS is using a SAT solver. So it might be
necessary to also change the algorithms.
Are you sure that you can devote enough time to achieve real advantages?
Any such effort should start by profiling to see where the time
actually goes - don't assume the SAT solver is eating it all. My
casual pokes at it showed a lot of time going into manipulating the
on-disk metadata, with a lot of short write() syscalls instead of
using buffered I/O - but that may not have been a representative
sampling of where the time goes.
You are confirming what I wrote about the need to change algorithms. The
SAT solver was just an example. Alas changing this doesn't seem to be so
easy without changing IPS requirements.
Andreas
_______________________________________________
openindiana-discuss mailing list
[email protected]
https://openindiana.org/mailman/listinfo/openindiana-discuss
