P. Van Roy S. Haridi and . Smolka. An overview of the design of Distributed Oz. In Proceedings of the Second International Symposium on Parallel Symbolic Computation (PASCO '97), pages 176{ 187, Maui, Hawaii, USA, July 1997. ACM Press.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... to a powerful model [Smolka, 1995] which allows for the integration of logic (constraint) functional, and object oriented programming, and also for smoothly passing from sequential to distributed programming, as this model adapts nicely to both these types of implementations [Mehl et al. 1995] [Haridi et al. 1997]. We use the CCP metaphor fair interleaving of constraint based computation processes and the logic characterization of attribute value matrices [Smolka, 1992] to decompose the HPSG ID schemata into quasi independent components. For instance, the ID schema 3 eventually combined ....

Haridi, S., Roy, P. V., and Smolka, G. (July 1997). An overview of the design of Distributed Oz. In Proceedings of PASCO '97, Maui, Hawaii.

....the system to be developed in a single, homogeneous, framework, and makes the distribution more transparent to the programmer. Functional languages potentially o er bene ts for small scale distributed programming, and several have been developed, e.g. Kali Scheme [CJK95] Facile [GMP89] OZ [HVS97] Concurrent Clean [PV98] and Pict [PT97] They allow high level distributed programming, e.g. capturing common patterns of distribution as higher order functions. Functional languages provide type safety within the constraints of a sophisticated, e.g. higher order and polymorphic, type system. ....

....have no explicit processes or threads but execute on multiple processors, e.g. HPF [For97] NESL [Ble96] These languages are not suitable for distribution, but have parallelism implicit in control or data structures. Shared name space languages support threads, but not processes, e.g. OZ [HVS97] Oblique [CJK95] GpH [THLP98] and Linda [GC92] Distributed name space languages support processes but not threads, e.g. Facile [GMP89] PICT [PT97] and languages based on communications libraries like C with MPI [The97] Shared and distributed name space languages support both threads ....

[Article contains additional citation context not shown here]

S. Haridi, P. Van Roy, and G. Smolka. An Overview of the Design of Distributed Oz. In 2nd Intl. Symposium on Parallel Symbolic Computation (PASCO 97), New York, USA, 1997.

....object oriented, and logic styles. It provides a variety of primitives for distribution and fault tolerance. Also much work has gone into planning and providing semantics for fault prone systems [Van99] Essentially it is the use of exceptions that allows an OZ programmer to deal with faults[HVS97, HVBS98] Lazy error detection, by handlers, enables the management of synchronous exceptions. Eager error detection, by watchers, enables the management of asynchronous exceptions. Fault tolerance is provided throughout the RTS down to the level of distributed garbage collection. 2.3.3 ....

S. Haridi, P. Van Roy, and G. Smolka. An overview of the design of Distributed Oz. In Proceedings of the Second International Symposium on Parallel Symbolic Computation (PASCO '97), pages 176-187, Maui, Hawaii, USA, July 1997. ACM Press.

....solving algorithms. However, concurrency brings important new challenges in many areas. As mentioned before, an important one from the implementation point of view is developing e ective analysis and optimization techniques. Distributed concurrent constraint systems are currently being worked on [7, 25], distribution and application development libraries are being o ered (e.g. 19, 8] and network and WWW applications are being reported [45] CP is a promising foundation for many aspects of the next generation of distributed systems, but it still remains as a challenge to develop simple, ....

S. Haridi, P. VanRoy, and G. Smolka. An Overview of the Design of Distributed Oz. In Proc. of the 1996 JISCLP Workshop 15 on Multi-Paradigm Logic Programming. T.U.Berlin, September 1996.

....search through databases which the agent simply is not allowed to access or if constraint satisfaction is too complex to be solvable by the agent without violating available time cost constraints. Available work that is relevant w.r.t. the distribution of the constraint handling process is, e.g. [26, 48, 49, 59, 60]. Condition 3. Agents must be able to reason about their constraints, and to involve other agents into this reasoning process whenever necessary. Following the argumentation in [32] see also [11] for earlier considerations on this subject) this kind of reasoning must be quantitative in nature, ....

S. Haridi, P. Van Roy, and G. Smolka. An overview of the design of Distributed Oz. In Proceedings of the Second International Symposium on Parallel Symbolic Computation (PASCO'97), pages 176-187, 1997.

....our similar Jinni engine [23] this implementation scenario shows that execution models going beyond Prolog s original LD resolution model are relatively easy to express in our framework. 5 Related work Similar to our Answer Sources, engine constructs [18] have been part of systems like Oz [10, 27] and have been used in the past for encapsulated search arguably more flexible than Prolog s fixed search mechanism. Besides the fact that our Answer Sources are instances of Fluents which provide a generic interaction mechanism with external stateful objects, independently of their execution ....

Seif Haridi, Peter Van Roy, and Gert Smolka. An Overview of the Design of Distributed Oz. In Proceedings of the Second International Symposium on Parallel Symbolic Computation (PASCO '97), pages 176--187, Maui, Hawaii, 1997. ACM Press.

....In particular, it discourages the implementation of multi agent systems where it is essential that all the participants can communicate on equal terms. For this reason, some LP systems utilise the Internet as their underlying communication layer. Two languages in this category are Distributed Oz [Haridi and Van Roy 1996] and April [McCabe and Clark 1995] Both use message passing and have the ability to move code between machines. April is not strictly speaking a LP language, but has borrowed ideas from LP, and its macro language can be used to support more Prolog like behaviour [Clark et al. 1996] SICStus ....

Haridi, S., and Van Roy, P. 1996. "An Overview of the Design of Distributed Oz", In Proc.

....of the low level handling of transport layers etc. By defining the behavior of different data structures, the network is still there, but on a higher level. The key concepts are network awareness and network transparency . The idea behind the distribution model of DML comes from Mozart [HVS97, HVBS98, VHB 97] Java s design goals include distributed programming in heterogenous networks, so we can hope to find a simple way of implementing distribution in DML. Java Remote Method Invocation (RMI) provides the mechanism by which servers and clients may communicate and pass information ....

....RMI RMI URL protocol URL protocol RMI Figure 7.1: RMI distributed computation model. 7.3 Distributed Semantics of DML To use the distribution correctly a semantics has to be specified for the primitive values. The semantics of DML tries to imitate the semantics of distributed Oz (see [HVS97] A summary of the semantics is depicted in Figure 7.2. Furthermore, while Java speaks in terms of clients and servers, the DML distribution model doesn t make such a distinction, we simply talk about sites. The home site of a value V is the site, where the value was created; other sites that ....

Seif Haridi, Peter Van Roy, and Gert Smolka. An Overview of the Design of Distributed Oz. In Proceedings of the Second International Symposium on Parallel Symbolic Computation (PASCO '97), pages 176--187, Maui, Hawaii, USA, July 1997. ACM Press.

No context found.

Seif Haridi, Peter Van Roy, and Gert Smolka. An Overview of the Design of Distributed Oz. In Proceedings of the Second International Symposium on Parallel Symbolic Computation (PASCO '97), pages 176--187, Maui, Hawaii, USA, July 1997. ACM Press.

.... in the context of logic programming [Robinson 1965; Warren 1977] They remain an essential part of logic programming and constraint programming systems [Van Roy 1994; Jaffar and Maher 1994] In the context of the Distributed Oz project, we have come to realize their usefulness to distribution [Haridi et al. 1997; Smolka et al. 1995] Logic variables express dependencies between computations without imposing an execution order. This property can be exploited in distributed computing: Two basic concerns in distributed computing are latency tolerance and thirdparty independence. We say a program is ....

....released in January 1999 [Mozart Consortium 1999] Mozart contains an optimized version of the on line DU algorithm. Distributed Oz, also known as Oz 3, conservatively extends Oz 2 to allow an efficient distributed network transparent implementation [Haridi et al. 1998; Van Roy et al. 1997; Haridi et al. 1997]. Oz 2 has a robust centralized implementation that was officially released in February 1998 [DFKI Oz 1998] Oz 3 keeps the same language semantics as Oz 2 and extends it with support for mobile computations, open distribution, component based programming, and orthogonal failure detection and ....

Haridi, S., Van Roy, P., and Smolka, G. 1997. An overview of the design of Distributed Oz. In the 2nd International Symposium on Parallel Symbolic Computation (PASCO 97). ACM.

....by other failures, e.g. if a site is unreachable for a long period of time. The detection variables and the rule S.3 reflect that phenomenon. 4 Network behavior. The network model has been designed to correspond to the Reliable Message Layer of the platform, which is currently built on TCP [HVS97] Yet it is surprising that a socalled reliable protocol can still result in message loss. But this is due to the way that TCP communicates with the source and destination sites. The network neighbourhoods net(S) model the buffers at the sending and receiving sites. What TCP guarantees is the ....

....the state pointer. As we said in the introduction, the state pointer is mobile, which means that it moves from site to site, so that pointer updates are always performed locally. 3. 1 Access structure and thread interface The distribution in the Mozart platform follows a general scheme [HVS97, HVBS98] Each site has a node called a proxy, and one special site has a node called the manager ; those nodes are called the access structure of the mutable pointer. Any other node accesses the pointer via its local proxy. The protocol manages the movements of the state pointer from proxy to ....

Seif Haridi, Peter Van Roy, and Gert Smolka. An overview of the design of Distributed Oz. In the 2nd International Symposium on Parallel Symbolic Computation (PASCO 97 ), New York, July 1997. ACM.

....A site in the model is a Mozart emulator on a computer, and the site fails when the emulator crashes. Moreover, the system is able to inform other sites that a site has failed. The network model has been designed to correspond to the Reliable Message Layer of the platform, which is built on TCP [6]. Section 2.2 may seem surprising to the reader, that a so called reliable protocol (TCP) can still result in message loss. But this is due to the way that TCP communicates with the source and destination sites. The network is not global at all, in fact it consists only of network ....

Seif Haridi, Peter Van Roy, and Gert Smolka. An overview of the design of Distributed Oz. In the 2nd International Symposium on Parallel Symbolic Computation (PASCO 97 ), New York, July 1997. ACM.

....record, procedure and thread. Anything else in Oz can be defined in terms of these basic entities. For instance, objects are defined in that way. Distributed Oz refines the semantics of Oz by attaching every entity to a site, and allowing some communication between certain remote entities [HVBS98, HVS97] Moreover, site failures and network problems are taken into account in the definition of the semantics. So each entity has a distributed behavior and a fault behavior. The entity that we focus on in this document, is the cell . Basically, a cell is a shared mutable pointer, and it is the basic ....

....to determine whether that site is alive, until the network becomes active again. That is the reason which we introduce detection variables for. 12 Network behavior. The network model has been designed to correspond to the Reliable Message Layer of the platform, which is currently built on TCP [HVS97] Yet it is surprising that a socalled reliable protocol can still result in message loss. But this is due to the way that TCP communicates with the source and destination sites. The network neighbourhoods net(S) model the buffers at the sending and receiving sites. What TCP guarantees is the ....

[Article contains additional citation context not shown here]

Seif Haridi, Peter Van Roy, and Gert Smolka. An overview of the design of Distributed Oz. In the 2nd International Symposium on Parallel Symbolic Computation (PASCO 97 ), New York, July 1997. ACM.

....level, they help make network transparent distribution practical. 1 Introduction Logic variables were first studied in the context of logic programming. They remain an essential part of logic programming and constraint programming systems [33, 16] In the context of the Distributed Oz project [12, 29], we have come to realize their usefulness to distribution. Logic variables express dependencies between computations without imposing an execution order. This property can be exploited in distributed computing: ffl Two basic problems in distributed computing are latency tolerance and third party ....

....of the on line DU algorithm, called the Oz algorithm, is part of the Mozart system, which implements the Distributed Oz language [1] An official release of Mozart is planned for late 1998. Distributed Oz, also known as Oz 3, is an efficient distributed network transparent extension to Oz 2 [11, 35, 12]. Oz 3 extends Oz 2 with support for mobile computations and orthogonal failure detection and handling within the language. Oz 2 has a robust centralized implementation that was officially released in February 1998 [5] Oz 2 programs are portable to Oz 3 almost immediately. Acknowledgements This ....

Seif Haridi, Peter Van Roy, and Gert Smolka. An overview of the design of Distributed Oz. In the 2nd International Symposium on Parallel Symbolic Computation (PASCO 97). ACM, July 1997.

....level, they help make network transparent distribution practical. 1 Introduction Logic variables were first studied in the context of logic programming. They remain an essential part of logic programming and constraint programming systems [36, 18] In the context of the Distributed Oz project [14, 31], we have come to realize their usefulness to distribution. Logic variables express dependencies between computations without imposing an execution order. This property can be exploited in distributed computing: ffl Two basic problems in distributed computing are latency tolerance and third party ....

....version of the on line DU algorithm, called the Oz algorithm, is part of the Mozart system, which implements the Distributed Oz language [1] An official release of Mozart is planned for late 1998. Distributed Oz, also known as Oz 3, is an efficient distributed networktransparent extension to Oz 2 [13, 38, 14]. Oz 2 is the successor to Oz 1 [29] which was designed for fine grained concurrency and implicit exploitation of parallelism. Oz 2 abandons this model in favor of explicit control over concurrency by means of a thread creation construct. Oz 2 has a robust centralized implementation that was ....

Seif Haridi, Peter Van Roy, and Gert Smolka. An overview of the design of Distributed Oz. In the 2nd International Symposium on Parallel Symbolic Computation (PASCO 97). ACM, July 1997.

....approach. 1.4 Mobility control and state The distributed semantics extends the language semantics with mobility control. In general terms, mobility control is the ability for stateful entities to migrate between sites or to remain stationary at one site, according to the programmer s intention [25]. The programmer can use mobility control to program the desired network communication patterns in a straightforward way. We define an entity to be a basic data item of the language, such as an object, a procedure, or a record. One use of mobile objects is as a cache to reduce network latency. ....

Seif Haridi, Peter Van Roy, and Gert Smolka. An overview of the design of Distributed Oz. In 2nd International Symposium on Parallel Symbolic Computation (PASCO 97). ACM, July 1997.

....logical concurrency in the application rather than to increase potential parallelism. 4 Distributed Oz Distributed Oz has the same language semantics as Oz. Distributed Oz separates application functionality from distribution structure by defining a distributed semantics for all language entities [52, 51, 17, 18]. The distributed semantics extends the language semantics to take into account the notion of site. It defines the network operations invoked when a computation is partitioned on multiple sites. We classify the language entities into three basic types (see Table 3) ffl Stateless entities are ....

Seif Haridi, Peter Van Roy, and Gert Smolka. An overview of the design of Distributed Oz. In Proceedings of the Second International Symposium on Parallel Symbolic Computation (PASCO '97), pages 176--187, Maui, Hawaii, USA, July 1997. ACM Press.

....approach. 1.4 Mobility Control and State The distributed semantics extends the language semantics with mobility control. In general terms, mobility control is the ability for stateful entities to migrate between sites or to remain stationary at one site, according to the programmer s intention [Haridi et al. 1997]. The programmer can use mobility control to program the desired network communication patterns in a straightforward way. For example, to reduce network latency a mobile object can behave as a state cache. This is illustrated by the shared graphic editor of Section 2. By stateful entities we mean ....

Haridi, S., Van Roy, P., and Smolka, G. 1997. An overview of the design of Distributed Oz. In the 2nd International Symposium on Parallel Symbolic Computation (PASCO 97 ). ACM, New York.

....operations 1 are predictable, which gives the programmer the ability to manage network communications. Mobility control is part of the distributed semantics. In general terms, mobility control is the ability for stateful entities to migrate between sites or to remain stationary at one site [14]. The programmer can use mobility control to program the desired network communication patterns in a straightforward way. For example, one use of mobile objects is as a cache to reduce network latency. This is illustrated by the shared whiteboard application in Section 2. By stateful entities we ....

Seif Haridi, Peter Van Roy, and Gert Smolka. An Overview of the Design of Distributed Oz. To appear in PASCO '97, July 1997. Available at http://www.info.ucl.ac.be/people/PVR/distribution.html.

No context found.

P. Van Roy S. Haridi and . Smolka. An overview of the design of Distributed Oz. In Proceedings of the Second International Symposium on Parallel Symbolic Computation (PASCO '97), pages 176{ 187, Maui, Hawaii, USA, July 1997. ACM Press.

No context found.

P. Van Roy S. Haridi and . Smolka. An overview of the design of Distributed Oz. In Proceedings of the Second International Symposium on Parallel Symbolic Computation (PASCO '97), pages 176-187, Maui, Hawaii, USA, July 1997. ACM Press.

No context found.

S. Haridi, P. VanRoy, and G. Smolka. An Overview of the Design of Distributed Oz. In Proc. of the 1996 JISCLP Workshop on Multi-Paradigm Logic Programming. T.U.Berlin, September 1996.

No context found.

S. Haridi, P. Van Roy, and G. Smolka. An overview of the design of distributed oz. In PASCO'97, july 1997. available at http://www.info.ucl.ac.be/people/PVR/distribution.html.

No context found.

S. Haridi, P. Van Roy, and G. Smolka. An Overview of the Design of Distributed Oz. In 2nd Intl. Symposium on Parallel Symbolic Computation (PASCO 97), New York, USA, 1997.

No context found.

S. Haridi, P. Van Roy, and G. Smolka. An Overview of the Design of Distributed Oz. In 2nd Intl. Symposium on Parallel Symbolic Computation (PASCO 97), New York, USA, 1997.

No context found.

S. Haridi, P. Van Roy, and G. Smolka. An Overview of the Design of Distributed Oz. In 2nd Intl. Symposium on Parallel Symbolic Computation (PASCO 97), New York, USA, 1997. 12

No context found.

S. Haridi, P. Van Roy, and G. Smolka. An overview of the design of Distributed Oz. In Proceedings of the Second International Symposium on Parallel Symbolic Computation (PASCO '97), pp. 176--187, Maui, Hawaii, USA, July 1997. ACM Press.

No context found.

Haridi, S.,van Roy, P. & Smolka, G. 1997 An Overview of the Design of Distributed Oz, PASCO 97 | International Symposium on Parallel Symbolic Computation, New York, USA.

No context found.

S. Haridi, P. Van Roy, and G. Smolka. An overview of the design of Distributed Oz. In Proceedings of the Second International Symposium on Parallel Symbolic Computation (PASCO'97), pages 176--187, 1997.

No context found.

S. Haridi, P. Van Roy, and G. Smolka, "An Overview of the Design of Distributed Oz", in Proc. 2nd Intl. Symposium on Parallel Symbolic Computation (PASCO 97), ACM, New York, USA, 1997.

No context found.

Seif Haridi, Peter Van Roy, and Gert Smolka. An overview of the design of Distributed Oz. In M. M. T. Chakravarty, Y. Guo, and T. Ida, editors, Multi-Paradigm Logic Programming (Proceedings of the JICSLP'96 PostConference Workshop), number 96-26 in Forschungsbericht, pages 1324. Technische Universit#t Berlin, Fachbereich Informatik, 1996.

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