Pong, F., Browne, M., Aybay, G. Nowatzyk, A. and Dubois, M., "Verifying Distributed Directory-based Cache Coherence Protocols: S3.mp, a Case Study", Proc. of the First Int'l EUROPAR Conf., Aug. 1995, pp. 287-300.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....can be more difficult when working with sophisticated theorem proving heuristics (see [10] Informal reasoning is commonplace in many verification efforts. For example, model checking is typically applied to an abstraction of the design that was produced informally by a verification expert [24,41]. Although the absence of errors in the abstraction does not guarantee the correctness of the actual design, errors found in the abstraction can reveal errors in the actual design. Many theorem prover based verifications model functional units at the register transfer level; the gateand ....

F. Pong, A. Nowatzyk, G. Aybay, and M. Dubois. Verifying distributed directory-based cache coherence protocols: S3.mp, a case study. In Proc. EURO-Par '95 Parallel Processing, number 966 in Lect. Notes Comput. Sci., pages 287--300. Springer-Verlag, August 1995.

....in a design, the verification is worthwhile even if some steps are justified only informally. Informal reasoning is commonplace in many verification efforts. For example, model checking is typically applied to an abstraction of the design that was produced informally by a verification expert [10, 19]. Although the absence of errors in the abstraction does not guarantee the correctness of the actual design, errors found in the abstraction can reveal errors in the actual design. Many theorem prover based verifications model functional units at the register transfer level: the gate and ....

F. Pong, A. Nowatzyk, G. Aybay, and M. Dubois. Verifying distributed directorybased cache coherence protocols: S3.mp, a case study. In S. Haridi, K. Ali, and P. Magnusson, editors, Proc. EURO-Par '95 Parallel Processing, number 966 in Lect. Notes Comput. Sci., pages 287--300. Springer-Verlag, August 1995.

....can be extremely complicated, especially in the presence of various optimizations. It often takes much more time to verify their correctness than to design them, and the problem of their verification has gained considerable attention in recent years [ABM93,Arc87,Bro90,PD95,PD96a] PD96b,PD96c,PNAD95,SD95,HQR99,Del00] Formal methods provide the only way to avoid subtle errors in sophisticated protocols. This paper addresses the task of implementing CRF in its own right. As part of this task, we propose a cache coherence protocol, Cachet [SAR99a,She00] which is adaptive in the sense that it ....

Fong Pong, Andreas Nowatzyk, Gunes Aybay, and Michel Dubois. Verifying Distributed Directory-based Cache Coherence Protocols: S3.mp, a Case Study. In Proceedings of the European Conference on Parallel Computing, 1995.

....[Gra94] has a more general method based on abstractions, which has been applied to a distributed cache memory, but it requires substantial manual effort to complete the proof. The work described here is closely related to the methods by Lubachevsky [Lub84] Dijkstra [Dij85] and Pong et al. PD95, PNAD95] Lubachevsky verified a concurrent This research was supported by Semiconductor Research Corporation under contract 95 DJ 389 and by the Advanced Research Projects Agency through NASA grant NAG 2 891. program by collapsing all reachable states into a fixed number of metastates , in which ....

....There are six restrictions on the use of RepetitiveID. In spite of the restrictions, RepetitiveID can be used to model a wide range of systems, including bus based multiprocessor cache coherence protocols [PD95] network based cache coherence protocols with a central or distributed directory [PNAD95, DDHY92, LLG 90] Intuitively, our goal is to isolate the parts of the state corresponding to the replicated components into a single array indexed by the RepetitiveID type. If two components i and j have identical component states, we would like the transition rule to produce a successor ....

F. Pong, A. Nowatzyk, G. Aybay, and M. Dubois. Verifying distributed directorybased cache coherence protocols: S3.mp, a case study. EurPar'95.

....an operational but fairly abstract view of Sequential Consistency, and then design protocols that admit exactly those behaviors that are permitted by this operational model. 1. 2 Formal Verification The verification of cache coherence protocols has gained considerable attention in recent years [4, 22]. Most methods verify certain invariants for cache coherence protocols, and are based on state enumeration [13, 14] and symbolic model checking [6, 7, 20] which can check correctness of assertions by exhaustively exploring all reachable states of the system. For example, Stern and Dill [27] use ....

F. Pong, A. Nowatzyk, G. Aybay, and M. Dubois. Verifying Distributed Directory-based Cache Coherence Protocols: S3.mp, a Case Study. In EuroPar'95, 1995.

....any bugs that arise is extremely difficult. ffl The protocols can be modeled with finite state, making them amenable to automatic formal verification. ffl Several researchers have successfully used automatic formal verification to debug complicated, real cache coherence protocols (e.g. [13, 7, 4, 10, 15, 17, 3, 14]) Unfortunately, in addition to the growing body of verification of cache coherence protocol success literature is also a growing oral tradition of verification failure stories. In the hallways at conferences or via private email, verification researchers informally trade stories of failed ....

....relevant to the real design all of the bugs that occur while implementing the design are not modeled and hence cannot be caught by protocol level verification. Nevertheless, the fact that protocol level verification of cache coherence protocols can be very valuable has been made repeatedly [13, 7, 4, 10, 15, 17, 3, 14]. Another negative is that our methodology in applying protocol level formal verification was far from ideal [7] ffl Ideally, protocol level formal verification is done extremely early in the design cycle. In our case, we started protocol level verification as RTL coding was nearing completion. ....

[Article contains additional citation context not shown here]

F. Pong, A. Nowatzyk, G. Aybay, and M. Dubois. Verifying distributed directory-based cache coherence protocols: S3.mp, a case study. In International Conferenceon Parallel Processing, EuroPar '95, August 1995.

....unspecified value. Dijkstra [10] verified a ring network by representing classes of similar states in regular expressions. Clarke and Grumberg [6] verified an alternating bit protocol by constructing an invariant process that records only the existence of components in a certain state. Pong et al. [31, 32, 33] verified many cache coherence protocols by representing classes of similar states using a set of repetition constructors, recording only whether there are zero, exactly one, one or more, or zero or more components in a certain state. Although these approaches cannot be used to verify deadlock ....

....and optimizations are useful in many cases, but a designer may find it error prone to incorporate them manually into the abstraction mapping. In order to reduce the amount of user effort, in this paper we present techniques to automate an abstraction similar to that proposed by Pong et al. [31, 32, 33]. An extension is implemented to the Mur verification system [11, 13] which provides a high level language in which a designer can easily specify a system in its concrete form. A new datatype, called RepetitiveID, is used to represent the indices of the replicated components. The Mur compiler ....

[Article contains additional citation context not shown here]

F. Pong, A. Nowatzyk, G. Aybay, and M. Dubois. Verifying distributed directory-based cache coherence protocols: S3.mp, a case study. First International EURO-PAR Conference on Parallel Processing, 1995.

No context found.

Pong, F., Browne, M., Aybay, G. Nowatzyk, A. and Dubois, M., "Verifying Distributed Directory-based Cache Coherence Protocols: S3.mp, a Case Study", Proc. of the First Int'l EUROPAR Conf., Aug. 1995, pp. 287-300.

....faster, larger systems are designed and built, the complexity of cache protocols will continue to increase. It is becoming impractical to verify a cache protocol by hand or by random simulations because a random test sequence must be run indefinitely to enter all reachable states. In some studies [66, 76], it has been shown that simulations are not very reliable in practice and, thus, there is a need for more efficient and reliable tools. As a result of the above compelling facts, recent research has focused on techniques that can verify protocols completely. This paper provides a survey of ....

....states is theoretically insufficient in general. The agreement between cache states and data copies must be guaranteed throughout the expansion process, which means that data values must be modeled along with state transitions. Consider the example of figure 7 of a stale write back error found in [76]. Suppose that cache A has a dirty copy of the block, replaces it, and writes it back to the home node. In order to guarantee that the memory receives the block safely, cache A keeps a valid copy of the block until it receives an acknowledgment from memory. Meanwhile, cache B sends a request for ....

[Article contains additional citation context not shown here]

Pong, F., Nowatzyk, A., Aybay, G. and Dubois, M., "Verifying Distributed Directory-based Cache Coherence Protocols: S3.mp, a Case Study", Proc. of the First Int'l EURO-PAR Conf., Aug. 1995, pp. 287-300.

....[16] Verifying a system with increasing numbers of caches becomes rapidly impractical in terms of computation time and memory requirement. As protocols become more complex, it is not clear whether verifying a small scale system model can provide a reliable error coverage for all system sizes [29]. The SSM method is a state enumeration approach which addresses the state space explosion problem by taking advantage of the homogeneity of the system and by exploiting equivalences among global states. More precisely, the method exploits an abstraction relation among global states so that a ....

....[3] does not distinguish between the two ambiguous messages and counts on the presumed owner to ignore the forwarded request. However, in a large scale system with unpredictable network delays, intractable problems can be caused by the forwarded request if it is further outpaced by other messages [29]. 3.3 Coherence Messages To solve the problem of ambiguous write back messages, we need different message IDs for a write back caused by a replacement (DOxMR) or by an invalidation (DOxMU) Moreover, when the (presumed) owner receives the forwarded request after the replacement, it replies with ....

[Article contains additional citation context not shown here]

Pong, F., Nowatzyk, A., Aybay, G. and Dubois, M., "Verifying Distributed Directory-based Cache Coherence Protocols: S3.mp, a Case Study", Proc. of the First Int'l EURO-PAR Conf., Aug. 1995, pp. 287-300. 36

....[15] Verifying a system with increasing numbers of caches becomes rapidly impractical in terms of computation time and memory requirement. As protocols become more complex, it is not clear whether verifying a small scale system model can provide a reliable error coverage for all system sizes [25]. Recently, we have introduced a new approach called SSM (Symbolic State Model) to address the state space explosion problem [23] and we have applied it to simple snoopy protocols on a single bus [2] SSM is a general framework for the verification of systems composed of homogeneous, communicating ....

....was also identified in [2] The solution suggested in [2] counts on the presumed owner to ignore the forwarded request. However, in a large scale system with unpredictable network delays, intractable problems can be caused by the forwarded request if it is further outpaced by other messages [25]. FIGURE 1. Directory state transitions to synchronize owner and memory. To solve the problem of ambiguous write back messages, we use different message IDs for a cache write back caused by a replacement (DOxMR) or by an invalidation (DOxMU) see table 1) Moreover we add two transient states to ....

[Article contains additional citation context not shown here]

Pong, F., Nowatzyk, A., Aybay, G. and Dubois, M., "Verifying Distributed Directory-based Cache Coherence Protocols: S3.mp, a Case Study", Proc. of the First Int'l EURO-PAR Conf., Aug. 1995, pp. 287-300.

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