W dniu 2 października 2008 19:56 użytkownik Kish Shen <kisshen_at_cisco.com> napisał: > Wit Jakuczun wrote: >> >> On 1 Paź, 19:47, neng.z..._at_gmail.com wrote: >> >>> but the CLP(FD) system is slightly slower. >>> >> Not sligthly but significantly! It stops me >> from using a new version of ECLiPSe in >> production system. >> >> There are however few nice features. The one >> I most like is source-based debugger. >> >> Best regards, >> Wit Jakuczun > > Hi Wit, > > Can you give me some details? I am surprised that there is significant > performance differences between 5.11 and 6.0 -- in almost all the tests we > have ran 6.0 was slightly faster. The thing that is slower is the > compilation time itself -- do you do a lot of compiling? > Compiling is not important. It is done once during initialisation phase of our module. We have developed a solver (using CLP) for tariff optimisation task (for a telecommunication company). The solver is based mainly on element constraint but there are also our global constraints specific to the problem and a lot of binary and real-value variables. The code has approx 2,5K lines of code divided into a few modules. Exemplary timings: Version: 5.10 6.0 0.87s 1.58s 0.13s 0.22s 0.13s 0.22s The times are averaged and the difference is statistically significant. In both cases I used the same data and have not changed any compilation flags. I used 5.10#127 and 6.0#42 versions of ECLiPSe. If you need any more details, please ask. Best regards -- [ Wit Jakuczun w.jakuczun_at_wlogsolutions.com ] [ WLOG Solutions http://wlogsolutions.com ]Received on Fri Oct 03 2008 - 06:29:26 CEST
This archive was generated by hypermail 2.2.0 : Thu Feb 02 2012 - 02:31:58 CET