U. Stern and D. L. Dill. Parallelizing the Mur# verifier. In O. Grumberg, editor, Computer-Aided Verification, 9th International Conference, volume 1254 of Lecture Notes in Computer Science, pages 256--267. Springer, June 1997. Haifa, Israel, June 22-25.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....algorithms is similar: the state graph is partitioned among the network nodes, i.e. each network node owns a subset of the state space. The differences are in the way the state space is partitioned (partition function) Probabilistic techniques to partition the state space have been used e.g. in [LS99, UD97, BBS01], and a technique which exploits some structural properties derived from the verified formula has been proposed in[BBv02] The model checking algorithm running on each network node has thus access only to a part of the entire system. Depending on the type of the algorithm it communicates with ....

U.Stern and D. L. Dill. Parallelizing the mur# verifier. In O. Grumberg, editor, Proceedings of Computer Aided Verification (CAV '97), volume 1254 of LNCS, pages 256--267. Springer-Verlag, 1997.

.... for 13 employing guided search heuristics [10] The partial order reduction can be employed more directly in the case of the simple DFS than in the case of the nested DFS [17] For terminal automata the algorithm is not tied up with Pcomplete DFS [25] and hence it allows for better distribution [29]. Some of these benefits were experimentally demonstrated by Edelkamp, Lafuente and Leue [10] They extended the model checker SPIN by a non emptiness algorithm which to a certain extent takes the type of an automaton into consideration. 4.2 Symbolic Algorithms Symbolic algorithms work with sets ....

U. Stern and D.L. Dill. Parallelizing the Mur# verifier. In Proc. Computer Aided Verification, volume 1254 of LNCS, pages 256--267. Springer, 1997.

....by parallel environment. Powerful parallel computers can be build of Networks Of Workstations (NOW) Thanks to various message passing interfaces (e.g. PVM, MPI) a NOW appears from the outside as a single parallel computer with a huge amount of memory. Reports by several independent groups ([33, 28, 17, 4, 3]) have confirmed the usefulness of distributed algorithms for the state space generation and reachability analysis. Methods for distributing LTL and CTL model checking have been presented in [1, 2, 8] and [6] respectively. However, until today not much effort has been taken to consider distributed ....

.... some of the regularity in the space and allow to verify systems with extremely large number of states, many orders of magnitude larger than could be handled by the explicit algorithms [11] Nevertheless, there are applications where explicit model checkers outperform the others, for examples see [33, 24, 25, 14] Thank to the fact that symbolic algorithms search the graph in a manner where the order in which vertices are visited is not crucial, these algorithms are directly parallelizable. On the other hand, the distribution of the BDD data structure is rather complicated. A parallel reachability ....

[Article contains additional citation context not shown here]

U. Stern and D.L. Dill. Parallelizing the Mur# verifier. In Proc. Computer Aided Verification, volume 1254 of LNCS, pages 256--267. Springer, 1997.

....algorithms is similar: the state graph is partitioned among the network nodes, i.e. each network node owns a subset of the state space. The di#erences are in the way the state space is partitioned (partition function) Probabilistic techniques to partition the state space have been used e.g. in [10,14,1], and a technique which exploits some structural properties derived from the verified formula has been proposed in [2] The model checking algorithm running on each network node has thus access only to a part of the entire system. Depending on the type of the algorithm it communicates with other ....

U.Stern and D. L. Dill, Parallelizing the mur# verifier, in: O. Grumberg, editor, Proceedings of Computer Aided Verification (CAV '97), LNCS 1254 (1997), pp. 256--267.

....power of such systems also helps in e#ectively reducing model checking time. Distributed algorithms for state space (and transition relation) generation from a high level system description using multiprocessor systems or a cluster of workstations have been reported recently, e.g. in [5,19,12,10,15]. All these papers, however, focus just on the generation of state spaces. The paper [3] studies parallel model checking of LTL, which is a di#erent kind of problem. In the current paper, we extend our previous work [12] and develop e#cient (in terms of computation and communication) algorithms ....

Stern, U. and D. Dill, Parallelizing the Mur# verifier, in: O. Grumberg, editor, Computer Aided Verification, Lecture Notes in Computer Science 1254 (1997), pp. 256--267.

....Our key observation is that we can omit this step yielding a simple parallel algorithm. Note that the game graph is also a Boolean graph and that our algorithm has similarities with the ones of [1,18] Until today, not much e#ort has been taken to consider parallel modelchecking algorithms. In [20], a parallel reachability analysis is carried out. The distribution of the underlying structure is similar to the one presented here. But their algorithm is not suitable for model checking temporal logic formulae. 13,22,2] present parallelised data structures which employ further computers within ....

....parallel. The idea of our parallel algorithm is that all processors are working in parallel on one component, whereas the components are treated one by one. Distributing the game graph. We employ a somehow standard approach distributing and constructing a (component of the) game graph in parallel [20,4,5]. As a data structure, we employ adjacency lists. We need also links to the predecessor as well as to the successor of a node for the labelling algorithm. A component is constructed in parallel by a typical breadth first strategy. Given anodeq, determine its successors q 1 , q n . To obtain a ....

U. Stern and D. L. Dill. Parallelizing the Mur# verifier. In O. Grumberg, editor, Computer-Aided Verification, 9th International Conference, LNCS 1254, p. 256--

....define an optimal search order in this sense and what the complexity of calculating it is. Such an optimal search order might in general make tools like Kronos and Uppaal more space e#cient. The close to linear speedups obtained were not very surprising since similar results have been obtained in [138] when constructing a distributed version of the model checker Mur#. Our approach to distributing Uppaal was inspired by the approach in [138] and the abstract algorithms of the two tools are basically the same, so we were hoping for as good results as in [138] Inclusion between states is not used ....

....tools like Kronos and Uppaal more space e#cient. The close to linear speedups obtained were not very surprising since similar results have been obtained in [138] when constructing a distributed version of the model checker Mur#. Our approach to distributing Uppaal was inspired by the approach in [138] and the abstract algorithms of the two tools are basically the same, so we were hoping for as good results as in [138] Inclusion between states is not used in Mur#, therefore the number of states searched is not influenced by the search order. The only di#erence between the two approaches is the ....

[Article contains additional citation context not shown here]

U. Stern and D. L. Dill. Parallelizing the Mur# verifier. In O. Grumberg, editor, Proc. of the 9th Int. Conf. on Computer Aided Verification, volume 1254 of Lecture Notes in Computer Science, pages 256--267. Springer--Verlag, June 1997. Haifa, Isreal, June 22-25.

....into the main memory of a single machine. Several Spin specific partition functions are experimented, the most advantageous one being a function that takes into account only a fraction of the state vector. Another distributed state enumeration algorithm has been implemented in the Mur verifier [34]. The speedups obtained are close to linear and the hash function used for state space partition provides a good load balancing. However, experimental data reported concerns relatively small state spaces (approximatively 1.5 M states) on a 32 node UltraSparc Myrinet network of workstations. There ....

....results haveshown that in this case there is almost no termination detection overhead, twowaves being always sufficient. This distributed termination detection scheme seems to use less messages than the centralized termination detection schemes used in the parallel versions of Spin [26] and Mur [34], which in all cases require several broadcast message exchanges between a coordinator machine and all other machines. 3.2 Merging of partitioned LTSs into monolithic LTSs After constructing a collection of N Bcg files representing a partitioned Lts byusingthe Distributor algorithm, the next ....

[Article contains additional citation context not shown here]

U. Stern and D. Dill. Parallelizing the Mur' Verifier. In Computer AidedVerification, volume 1254, pages 256--267. Lecture Notes in Computer Science, SpringerVerlag, 1997.

.... the work by Holzmann [16] Wolper and Leroy [30] and Stern and Dill [24] The latter two authors recently also described a distributed state space exploration approach very similar to the one presented by Caselli et al. Ciardo et al. and Knottenbelt et al. yielding very good speed ups as well [25]. In the sections that follow we will present our recent results in this area. To ease the explanation of our work, we first present an example SPN in Section 2; it has been taken from the literature and has been used widely as benchmark SPN. Section 3 is devoted to the sequential generation of ....

....presented by Caselli et al. 6] Ciardo and 12 530 540 550 560 570 580 590 600 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0. 9 1 t degree of filling k = 7 Figure 6: The generation time (in seconds) for the FMS example with k = 7 for varying hash table degree of filling Nicol [8] and Stern and Dill [25] and works as follows (see also Figure 7) ffl all processors start their exploration program; ffl processor i with i = Z(Initial) starts to explore successor states; ffl upon generating a state s 0 from state s: the allocation for the new state s 0 is computed: j : Z(s 0 ) if j ....

U. Stern and D.L. Dill. Parallelizing the Mur' verifier. In O. Grumberg, editor, Computer Aided Verification, Lecture Notes in Computer Science 1254, pages 256--267. SpringerVerlag, 1997.

....the very large number of reachable states of most protocols. The second approach aims at exploring a given reachability graph in the most efficient manner, minimizing memory usage and runtime. Examples are bitstate hashing [4] hash compaction [15, 12] and parallelizing the state space search [14]. In this paper, we present a technique that reduces the main memory requirement of the state table maintained in explicit state enumeration. The state table eventually holds all reachable states of the protocol under verification unless an error is detected. In addition, the state table is ....

.... be used in other explicit state verification tools like SPIN [5] In addition, the algorithm is compatible with the two newer reduction techniques in Mur [8] reversible rules and repetition constructors, which were not yet available in the public release, and with the parallel version of Mur [14]. For checking liveness properties, the currently used algorithms require a (modified) depth first search of the state space. Since the new scheme is based on a breadth first search of the state space, it is not directly compatible with these algorithms. Two other recent techniques, the most ....

[Article contains additional citation context not shown here]

U. Stern and D. L. Dill. Parallelizing the Mur' verifier. In Computer Aided Verification. 9th International Conference, 1997.

....kinds of atomic transactions. It is much easier or trivial to prove important properties of the reduced model, such as the consistency of data at the user level, than the original protocol description. To check the abstraction automatically, we have run Parallel Mur on 32 Ultra Sparc processors [27]. For the protocol with 3 processing nodes and request reply message queues of size 5, the verifier explored 457,558 states in 126 seconds; for 4 processing nodes and queues of size 3, about 19 million states in 72 minutes. As expected, the number of states explored (also the usage of memory) is ....

Ulrich Stern and David L. Dill. Parallelizing the Mur' verifier. In Computer Aided Verification, 9th International Conference, CAV'97. Springer-Verlag, June 1997.

....the reachability analysis. Examples are bitstate hashing [9] hash compaction [28, 22] and using magnetic disk instead of main memory to store the state space [21] A preliminary version of this paper was published in the 9th International Conference on Computer Aided Verification, 1997 [23]. In this paper, we explore a third approach to improve explicit state enumeration: parallel processing. With the use of state and memory reduction techniques, runtime becomes a major limiting factor [28, 22] For example, when verifying complex protocols with the Mur verifier [7] using ....

U. Stern and D. L. Dill. Parallelizing the Mur' verifier. In Computer Aided Verification. 9th International Conference, pages 256--67, 1997.

.... Second, Mur currently implements a richer set of methods for increasing the size of the protocols that can be verified, including symmetry reduction [4] hash compaction [17] reversible rules [5] and repetition constructors [6] In addition, there is a parallel version of the Mur verifier [15]. Although available for internal use, the latter three techniques are not yet in the public Mur release. 2.2 The methodology In outline, we have analyzed protocols using the following sequence of steps: 1. Formulate the protocol. This generally involves simplifying the protocol by identifying ....

U. Stern and D. L. Dill. Parallelizing the Mur' verifier. Submitted for publication.

....systems. The second approach aims at exploring a given reachability graph in the most efficient manner, minimizing memory usage and runtime (both of which are limiting factors in verification) Examples are bitstate hashing [5] hash compaction [17, 14] and parallelizing the state space search [16]. In this paper, we describe a technique that reduces the main memory requirements of the state table maintained in explicit state enumeration. The state table eventually holds all reachable states of the system being verified unless an error is detected. In addition, the state table is typically ....

....machine with a huge main memory. The algorithm described could also be used in other explicit state verification tools like SPIN [6] In addition, the algorithm is compatible with all three state reduction techniques in Mur [8] with hash compaction, and with the parallel version of Mur [16]. The algorithm is also compatible with Peled s partial order method [11] which had been assumed to require depth first search, but was recently shown to also work with breadth first search [3] on which the new scheme is based. This recent finding suggests that other partial order methods might ....

[Article contains additional citation context not shown here]

U. Stern and D. L. Dill. Parallelizing the Mur' verifier. In Computer Aided Verification. 9th International Conference, pages 256--67, 1997.

....techniques. The three state reduction techniques in Mur are symmetry reduction [45] reversible rules [47] and repetition constructors [48] The algorithmic techniques in Mur are the techniques described in this thesis hash compaction [77, 80] a parallel version of the Mur verifier [81], and a version that allows using magnetic disk instead of main memory for storing the state space [75] Although all the techniques are available for internal use, currently only symmetry reduction and hash compaction are available in the public Mur release. 2.2 The Needham Schroeder Protocol ....

....Berkeley (SPARC20s connected via Myrinet) using generic active messages [16] as the message passing layer; later it was ported with little effort to an SP2 at IBM Watson. 1 This chapter is based on material first published in the 9th International Conference on Computer Aided Verification, 1997 [81]. In parallel Mur , the state table, which stores all reached protocol states, is partitioned over the nodes of the parallel machine. Thus, the table can be larger than on a single node. Each node maintains a work queue of unexplored states. When a node generates a new state, the owning node ....

U. Stern and D. L. Dill. Parallelizing the Mur' verifier. In Computer Aided Verification. 9th International Conference, pages 256--67, 1997.

No context found.

U. Stern and D. L. Dill. Parallelizing the Mur# verifier. In O. Grumberg, editor, Computer-Aided Verification, 9th International Conference, volume 1254 of Lecture Notes in Computer Science, pages 256--267. Springer, June 1997. Haifa, Israel, June 22-25.

No context found.

Stern, U., Dill, D.L.: Parallelizing the Mur# verifier. In Grumburg, O., ed.: Computer-Aided Verification, CAV '97. Volume 1254 of Lecture Notes in Computer Science., Haifa, Israel, Springer-Verlag (1997) 256--267

No context found.

U. Stern and D. L. Dill. Parallelizing the Mur# verifier. In O. Grumberg, editor, Computer-Aided Verification, 9th International Conference, volume 1254 of Lecture Notes in Computer Science, pages 256--267. Springer, June 1997. Haifa, Israel, June 22-25.

No context found.

Ulrich Stern and David Dill. Parallelizing the Mur' verifier. Formal Methods in System Design, 18(2):117--129, 2001. (Journal version of their CAV 1997 paper).

No context found.

Ulrich Stern and David Dill, "Parallelizing the Mur# verifier," Formal Methods in System Design, vol. 18, no. 2, pp. 117--129, 2001, (Journal version of their CAV 1997 paper).

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