Christian Schulte. Comparing trailing and copying for constraint programming. In Danny De Schreye, editor, Proceedings of the Sixteenth International Conference on Logic Programming, pages 275--289. The MIT Press, December 1999.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....node by incrementing the idx member of the parent. 5 Copying Based Search and Relative Addressing Note that in the previous section the same store is passed between solve one and child. Search is done entirely by trailing and backtracking, as in most constraint programming systems. Schulte [Sch99] shows that copying based search as employed by the Mozart system, combined with recomputation of spaces, can compete with the performance of trailing based systems. To study the performance of memory policies, it appears to be attractive to provide both copying and trailing in the same system. ....

....Memory Policy Relative addressing of variables and propagators in stores in Figaro, which is not present in Ilog Solver, is the the key feature that allows to use copying based search in addition to trailing. Both copying based and trailing based search can be combined with recomputation [Sch99] We hope that the exibility to use both memory policies leads to interesting, possibly adaptive, and more ecient combinations of the two memory policies and recomputation. Another memory management issue is the creation of nodes, which are in the presented simpli ed design not explicitly ....

Christian Schulte. Comparing trailing and copying for constraint programming. In Danny De Schreye, editor, Proceedings of the International Conference on Logic Programming, pages 275-289, Las Cruces, New Mexico, August 1999. The MIT Press, Cambridge, MA.

....does not have to be trailed. There exists also a runtime technique for preventing the multiple value trailing between two choice points: see for instance [11] However, this technique only works in the WAM scheme, because it introduces linear reference chains that PARMA does not allow. Finally, [14] and [20] also discuss the reconstruction of a state on backtracking by copying and recomputing techniques respectively. The context of those works is quite different, but the latter technique is naturally related to a hybrid technique mentioned later in this section. There is little room left ....

C. Schulte. Comparing trailing and copying for constraint programming. In D. De Schreye, editor, Proceedings of the Sixteenth International Conference on Logic Programming, pages 275--289, Las Cruces, NM, USA, Nov. 1999. The MIT Press.

....does not have to be trailed. There exists also a runtime technique for preventing the multiple value trailing between two choice points: see for instance [11] However, this technique only works in the WAM scheme, because it introduces linear reference chains that PARMA does not allow. Finally, [14] and [20] also discuss the reconstruction of a state on backtracking by copying and recomputing techniques respectively. The context of those works is quite di erent, but the latter technique is naturally related to a hybrid technique mentioned later in this section. There is little room left for ....

C. Schulte. Comparing trailing and copying for constraint programming. In D. D. Schreye, editor, Proceedings of the Sixteenth International Conference on Logic Programming, pages 275-289, Las Cruces, NM, USA, Nov. 1999. The MIT Press.

....a literal is guessed, two independent computations can be spawned, one which assumes the literal to be true and one that assumes the literal to be false. Exploitation of Vertical Parallelism shares the same roots as or parallelism for Prolog [10] and search parallelism in constraint programming [20, 17]. Recent studies have underlined the inherent complexity of maintaining the correct view of execution during parallel search [19] The overall design of the engine used for our experiments has been directly derived from the design of the engine used in the smodels system [15] In this paper we are ....

....approach to work sharing adopts a simpler approach then recomputation. Upon work sharing from agent A to B, the entire partial answer set existing in A is directly copied to agent B. The use of copying has been frequently adopted to support computation in constraint programming systems [20] as well as to support orparallel execution of logic and constraint programs [10] The partial answer set owned by A has an explicit representation within the agent A: it is completely described by the content of the trail stack. Thus, copying the partial answer set from A to B can be simply ....

C. Schulte. Comparing Trailing and Copying for Constraint Programming. In International Conference on Logic Programming, pages 275-289. MIT Press, 1999.

....(discrepancies) from the branching strategy. The most common implementation technique for exploration algorithms is trailing based backtracking, also inherited from Prolog. This technique works well for sequential depth rst search and is employed by CHIP, Ilog Solver OPL and Claire. Schulte [Sch99] showed that copying of constraint stores together with recomputation, which is employed by Mozart [Moz99] and has advantages for exploration algorithms other than sequential depth rst search, can be competitive with backtracking. In order to achieve combinatorial optimization as in scheduling, ....

Christian Schulte. Comparing trailing and copying for constraint programming. In Proceedings of the International Conference on Logic Programming, 1999. to appear.

.... Program 4 Backtracking Depth First Search choice solveone(store s,choice c) if (c = NULL) return c; mark m = s mark( try return solveone(s,c choose(s,0) catch (Failure) s backtrack(m) return solveone(s,c choose(s,1) is described by Schulte in [Sch97b] In [Sch99], Schulte analyses recomputation, a method for reducing the space consumption of copying based search, and compares several variants of recomputation with backtracking based search. In order to support copying recomputation, we allow to copy stores through the following operation. store ....

Christian Schulte. Comparing trailing and copying for constraint programming. In Proceedings of the International Conference on Logic Programming, 1999. to appear.

....catch (Failure) printf( no solution n ) 5 Copying Based Search and Relative Addressing Note that in the previous section the same store is passed between solveone and child. Search is done entirely by trailing and backtracking, as in most constraint programming systems. Schulte [Sch99] shows that copying based search as employed by the Mozart system, combined with recomputation of spaces, can compete with the performance of trailing based systems. To study the performance of memory policies, it appears to be attractive to provide both copying and trailing in the same system. ....

....Memory Policy Relative addressing of variables and propagators in stores in Figaro, which is not present in Ilog Solver, is the the key feature that allows to use copying based search in addition to trailing. Both copying based and trailing based search can be combined with recomputation [Sch99] We hope that the flexibility to use both memory policies leads to interesting, possibly adaptive, and more efficient combinations of the two memory policies and recomputation. Another memory management issue is the creation of nodes, which are in the presented simplified design not explicitly ....

Christian Schulte. Comparing trailing and copying for constraint programming. In Proceedings of the International Conference on Logic Programming, 1999. to appear.

....in tell(r,x, xl,yu 1) tell(r,y, xl 1,yu) if xu yl then dropProp(r,p) else ( end fun lt(r,x,y) prop(r, x,y] lt ) 6 Copying based Search Most programming languages and systems for constraint programming use a backtrackingbased approach as presented in the previous sections. Schulte [Sch99] shows that copyingbased search, combined with recomputation of spaces, can compete with the performance of backtracking based systems. Thus it appears to be attractive to provide both copying and backtracking in the same system. We show how our search abstractions can be modified to support ....

Christian Schulte. Comparing trailing and copying for constraint programming. In Proceedings of the International Conference on Logic Programming, 1999. to appear.

....to allow stopping infinite searches, and with the option to return solutions either directly or encapsulated in computation spaces (see Section 7. 4) Search implemented with spaces using strategies combining cloning and recomputation is competitive in time and memory with systems using trailing [92]. Using encapsulation, general purpose search can be used as a primitive to build more sophisticated searches. Parallel search. When provided with a list of machines, this will spread out the search process over these machines transparently. We have benchmarked realistic constraint problems on ....

....Spaces are a natural way to integrate search into a concurrent system. Spaces can be implemented e#ciently: on real world problems the Mozart 1.1. 0 implementation using copying and recomputation is competitive in time and memory use with traditional systems using trailing based backtracking [92]. Spaces are compositional, i.e. they can be nested, which is important for building well structured programs. This section defines computation spaces, the operations that can be performed on them (see Figure 5) and gives a few examples of how to use them to program search. The discussion in ....

Christian Schulte. Comparing trailing and copying for constraint programming. In Danny De Schreye, editor, Proceedings of the 1999.

....Systems, 2000 [129] Christian Schulte. Programming Deep Guard Concurrent Constraint Combinators. Practical Aspects of Declarative Languages, 2000 [130] Christian Schulte. Comparing Trailing and Copying for Constraint Programming. International Conference on Logic Programming, 1999 [126]. Christian Schulte. Programming Constraint Inference Engines. International Conference on Principles and Practice of Constraint Programming, 1997 [125] Christian Schulte. Oz Explorer: A Visual Constraint Programming Tool. International Conference on Logic Programming, 1997 [124] ....

Christian Schulte. Comparing trailing and copying for constraint programming. In De Schreye [28], pages 275--289.

....and memory use) with an asynchronous kill procedure to allow stopping in nite searches, and with the option to return solutions either directly or encapsulated in computation spaces (see Section 7. 4) Search implemented with spaces is competitive in time and memory with systems using trailing [66]. Using encapsulation, general purpose search can be used as a primitive to build more sophisticated searches. Parallel search. When provided with a list of machines, this will spread out the search process over these machines transparently. We have benchmarked realistic constraint problems on ....

....Spaces have the exibility needed for real world constraint problems and they can be implemented eciently: on real world problems the Mozart 1.1. 0 implementation using copying and recomputation is competitive in time and memory use with traditional systems using trailing based backtracking [66]. Spaces are a natural way to integrate search, which has a strong speculative component, into a concurrent system that interacts with its environment. We say an execution is speculative if it is not guaranteed to be part of the nal computation. Managing speculative computation is the essence ....

Christian Schulte. Comparing trailing and copying for constraint programming. In International Conference on Logic Programming (ICLP 99), pages 275-289. MIT Press, November 1999.

....6 8 10 2 0.6 3 1.9 4 2.9 5 4.5 6 4.7 0 2 4 6 8 10 2 0.3 3 0.8 4 1.5 5 2.0 6 2.8 (worker) worker) Fig. 10. Recomputation overhead. This means that the price paid for distributing work across the network is low. For the general performance of recomputation in Oz, see [17]. Exploration Overhead Exploration overhead occurs for branch and bound search and is due to the di#ering order in which solutions are found (see Section 3.4) Figure 11 shows the exploration overhead for Photo and MT 10. The exploration overhead is almost exclusively the cause for the speedup ....

Christian Schulte. Comparing trailing and copying for constraint programming. In Danny De Schreye, editor, Proceedings of the 1999 International Conference on Logic Programming, pages 275--289, Las Cruces, NM, USA, November 1999. The MIT Press.

....and memory use) with an asynchronous kill procedure to allow stopping infinite searches, and with the option to return solutions either directly or encapsulated in computation spaces (see Section 7. 4) Search implemented with spaces is competitive in time and memory with systems using trailing [62]. Using encapsulation, general purpose search can be used as a primitive to build more sophisticated searches. ffl Parallel search. When provided with a list of machines, this will spread out the search process over these machines transparently. We have measured realistic constraint problems on ....

....Spaces have the flexibility needed for real world constraint problems and they can be implemented efficiently: on real world problems the Mozart 1.1. 0 implementation using copying and recomputation is competitive in time and memory use with traditional systems using trailing based backtracking [62]. Spaces are a natural way to integrate search, which has a strong speculative component, into a concurrent system that interacts with its environment. We say an execution is speculative if it is not guaranteed to be part of the final computation. Managing speculative computation is the essence ....

Christian Schulte. Comparing trailing and copying for constraint programming. In International Conference on Logic Programming (ICLP 99), pages 275--289. MIT Press, November 1999.

No context found.

Christian Schulte. Comparing trailing and copying for constraint programming. In Danny De Schreye, editor, Proceedings of the Sixteenth International Conference on Logic Programming, pages 275--289. The MIT Press, December 1999.

No context found.

Schulte, C. 1999. Comparing Trailing and Copying for Constraint Programming. In Proceedings of the Sixteenth International Conference on Logic Programming, D. De Schreye, Ed. MIT Press, Las Cruces, NM, USA, 275--289.

No context found.

C. Schulte. Comparing trailing and copying for constraint programming. In ICLP'99.

No context found.

C. Schulte. Comparing trailing and copying for constraint programming. In ICLP'99.

No context found.

Christian Schulte. Comparing trailing and copying for constraint programming. In Danny De Schreye, editor, Proceedings of the International Conference on Logic Programming, pages 275-289, Las Cruces, New Mexico, August 1999. The MIT Press, Cambridge, MA.

No context found.

C. Schulte. Comparing trailing and copying for constraint programming. In Proceedings of the International Conference on Logic Programming, ICLP, pages 275--289. The MIT Press, 1999.

No context found.

Christian Schulte. Comparing Trailing and Copying for Constraint Programming. 999. Internet: http://ps.uni-sb.de/~schulte/papers/copying.ps.gz

No context found.

C. Schulte. Comparing trailing and copying for constraint programming. In D. D. Schreye, editor, Proceedings of the Sixteenth International Conference on Logic Programming, pages 275-289. MIT Press, November 1999.

No context found.

Christian Schulte. Comparing trailing and copying for constraint programming. In Proc. of ICLP'99, pages 275--289, Las Cruces, New Mexico, USA, November 29, December 4 1999.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC