### Table 2: Comparison with ILOG Solver and Ant-P

2001

"... In PAGE 4: ... Let us now compare with a constraint programming system (ILOG solver) and an ant colony optimization method (Ant-P solver), both timings (in seconds) are taken from [23] and divided by a factor 7 corresponding to the SPECint 95 ratio be- tween the processors. Table2 clearly show that adap- tive search is much more performant on this benchmark, which might not be very representative of real-life ap- plications but is a not-to-be-missed CSP favorite.... ..."

Cited by 25

### Table 1 presents computational results on a number of test problems. As constraint programming solver we have used ILOG Solver, version 6.0. The experiments were performed on a Pentium III 550MHz, 4GB RAM, with a time limit of 300 seconds per instance.

2006

"... In PAGE 2: ... Table1 : Computational results on a number of C TAEMS task structures. All instances are solved to optimality, unless the time limit (300s) has been reached.... ..."

Cited by 1

### Table 16. Number of FD variables managed in the magic sequences problem Ilog clp(fd) Oz ECLiPSe

in A Comparative Study of Eight Constraint Programming Languages over the Boolean and Finite Domains

"... In PAGE 25: ...ubsections 5.3.2 and 5.3.3, the robustness of each the systems was measured. For this, we used the magic sequences (N) programs (with garbage collection on) and measured the maximum value of N that each system could manage. Table16 gives... ..."

### Table 1. Program sizes

"... In PAGE 5: ...ally dynjava (http://koala.ilog.fr/djava/index.html) is a dy- namic java source code interpreter that executes programs written in dynamic java language. Some parameters of these test programs and the number of test cases we defined are presented in Table1 . The num- ber of program lines is the number of those lines for which bytecode instructions were generated, summarized for all... ..."

Cited by 1

### Table 2: Results on an industrial project scheduling problem 6 Conclusions We have shown how to combine operations research and arti cial intelligence, in particular constraint programming, in a way that preserves the best of both, i.e., we preserved the e ciency provided by operations research and the generality of approach o ered by constraint programming. We have shown that a good performance can be obtained on such a classical scheduling problem as the JSSP as well as on generalizations thereof, and on real-life scheduling problems as described in Section 5.3. In short, we think that in Ilog Schedule we have found a powerful combination of techniques that allows us to tackle a broad range of practical scheduling problems in an e cient way.

1995

"... In PAGE 10: ... As a result, the nal user wants to impose upper bounds on the two criteria, and wants the system to tell whether there exists a solution satisfying these upper bounds. Table2 reports the CPU times, in seconds, obtained for di erent values of the upper bounds of the two criteria. We compared two algorithms, one using arc-B-consistency and one using the edge- nder of Nuijten [1994].... In PAGE 10: ... Numbers in bold correspond to the upper bounds for which there is no solution. Table2 clearly shows that the edge- nder strongly outperforms arc-B-consistency.... ..."

Cited by 24

### Table 2: Results on an industrial project scheduling problem 6 Conclusions We have shown how to combine operations research and arti cial intelligence, in particular constraint programming, in a way that preserves the best of both, i.e., we preserved the e ciency provided by operations research and the generality of approach o ered by constraint programming. We have shown that a good performance can be obtained on such a classical scheduling problem as the JSSP as well as on generalizations thereof, and on real-life scheduling problems as described in Section 5.3. In short, we think that in Ilog Schedule we have found a powerful combination of techniques that allows us to tackle a broad range of practical scheduling problems in an e cient way.

1995

"... In PAGE 10: ... As a result, the nal user wants to impose upper bounds on the two criteria, and wants the system to tell whether there exists a solution satisfying these upper bounds. Table2 reports the CPU times, in seconds, obtained for di erent values of the upper bounds of the two criteria. We compared two algorithms, one using arc-B-consistency and one using the edge- nder of Nuijten [1994].... In PAGE 10: ... Numbers in bold correspond to the upper bounds for which there is no solution. Table2 clearly shows that the edge- nder strongly outperforms arc-B-consistency.... ..."

Cited by 24