Rapid Parameterized Model Checking of Snoopy Cache Coherence Protocols

We propose new, tractably (in some cases provably) efficient algorithmic methods for exact (sound and complete) parameterized reasoning about cache coherence protocols. For reasoning about general snoopy cache protocols, we introduce the guarded broadcast protocols model and show how an abstract history graph construction can be used to reason about safety properties for this framework. Although the worst case size of the abstract history graph can be exponential in the size of the transition diagram of the given protocol, the actual size is small for standard cache protocols as is evidenced by our experimental results. The framework can handle all 8 of the cache protocols in [19] as well as their split-transaction versions. We next identify a framework called initialized broadcast protocols suitable for reasoning about invalidation-based snoopy cache protocols and show how to reduce reasoning about such systems with an arbitrary number of caches to a system with at most 7 caches. This yields a provably polynomial time algorithm for the parameterized verification of invalidation based snoopy protocols. Our results apply to both safety and liveness properties. Finally, we present a methodology for reducing parameterized reasoning about directory based protocols to snoopy protocols, thus leveraging techniques developed for verifying snoopy protocols to directory based ones, which are typically are much harder to reason about. We demonstrate by reducing reasoning about a directory based protocol suggested by German [17] to the ESI snoopy protocol, a modification of the MSI snoopy protocol.

In this paper, we propose two different algorithms for Shannon Factor Graph (SFG) construction, which can be used for cut-less mapping, to improve the runtime, graph size and required memory size. The first SFG construction algorithm does not consider the nature of the nodes (constant one and zero, non-decomposed and shared nodes) while building the SFG, whereas the second algorithm finds out the nature of the nodes on-the-fly. We observed that the constant one and zero, Shared and non-decomposed nodes can be used at the time of SFG construction to minimize the runtime and graph size significantly and to make the graph semi-canonical. The theoretical analysis and experiments performed on the standard benchmark circuits show that, by finding the constant one and zero, shared and non-decomposed nodes on-the-fly reduces the graph size by a factor of 126 and the runtime by a factor of 5.5.

With the development of multi-core and many-core processor architecture, cache consistency protocols become increasingly complex, verification is also becoming more challenging. Regression verification runs through the whole process of chip design and is an indispensable part of chip design and verification. It takes a lot of time and resources to execute a gigantic set of test cases. In this paper, a test case priority ranking method based on multi-objective optimization is proposed, which is applied to the regression verification for RTL design of an original multi-core processor cache coherence protocol. Based on the detection rate of defect prediction and function coverage rate as the optimization objectives, the test case priority is ranked. The experimental results show that the proposed method can improve the effectiveness of simulation verification, reduce time and cost, find defects as soon as possible, improve the reliability of the chip, is of great theoretical significance and of engineering practical value.

We develop a specification methodology that documents and specifies a cache coherence protocol in eight tables: the states, events, actions, and transitions of the cache and memory controllers. We then use this methodology to specify a detailed, modern three-state broadcast snooping protocol with an unordered data network and an ordered address network that allows arbitrary skew. We also present a detailed specification of a new protocol called multicast snooping (Bilir et al., 1999) and, in doing so, we better illustrate the utility of the table-based specification methodology. Finally, we demonstrate a technique for verification of the multicast snooping protocol, through the sketch of a manual proof that the specification satisfies a sequentially consistent memory model.

Interest Rate Rules and Nominal Determinacy

A refined plastic hinge analysis method, known as one of the most effective and practical second-order inelastic analysis methods for steel frames, is able to evaluate the ultimate structural behavior of steel frames by considering the geometric and material nonlinearities. However, an appropriate advanced nonlinear solution technique has to be incorporated to help structural engineers perform the rational design of steel frames by predicting the ultimate strength and post-failure structural behavior accurately. In this study, a refined plastic hinge analysis method, combined with a work-increment-control solution technique with an iterative procedure in incremental loading steps, was presented in order to overcome the shortcomings of the load-increment-control solution techniques employed in previous studies. In the work-increment-control solution technique of present study, one convergence criterion refining the problem of using two convergence criteria in the conventional increment/iteration procedure of work-increment-control solution techniques and an automatic incremental algorithm calculating the load factor and the magnitude of incremental work for next incremental loading step were employed. To verify the accuracy and appropriateness of the present approach, three representative plane steel frames employed in previous studies were analyzed, and the analysis results were compared with those by other approaches. The present approach, that evaluated fairly accurately the load-displacement relationships, ultimate loads, plastic hinge numbers and locations, and the post-critical responses up to the formation of collapse mechanism of the plane steel frames, proved to be acceptable.

We have studied Open Data as a connected graph. Each data package is considered a vertex, and we studied the similarity graph induced by several different similarity measures. We analyzed the resulting similarity graph using different metrics to estimate its quality and informativeness. In order to cope with the size of the open data graph (over 6 billion edges), the graph constructions and analysis are done using a distributed computation framework, Apache Spark. The algorithms were implemented using the Spark resilient distributed data algebra, and executed on the Google Cloud Platform (GCP).

Cache coherence protocols of increasing complexities call for automated verification tools which are both efficient and reliable. Most current approaches can only verify protocols at a high level of abstraction and the model size is limited to a small number of interacting processes. By using a simple full-map directory scheme as example, we present a verification technique which is extremely efficient and is independent of the model size. Several non-obvious problems affecting the correctness of a protocol design are identified by the verification procedure.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

A theoretical and experimental study of a new algorithm for minimum cost flow in dynamic graphs

Meta Knowledge Graph is mainly a knowledge graph formed by the most basic concept entity and its synonymous relationship, which is the most direct entry to show the basic knowledge of the industry. However, due to the complexity of its data sources and the lack of data description specification, it has always been a research difficulty. Based on the analysis of big data characteristics of the power industry and SG-CIM data characteristics of the power company, this paper takes power basic terms as the entrance, crawls complete term entries from the network, obtains entities and synonyms by parsing content, and forms association relationship, and then uses graph database and programming tools to solve the technical problems of Meta Knowledge Graph storage and user interaction. It provides a practical method and technical scheme for the construction of electric power meta knowledge graph.

Snoop-based cache coherence protocols are typically used when multiple processor cores share memory through a common bus. It is well known, however, that these coherence protocols introduce an excessive power overhead. To help alleviate this problem, we propose an application-driven customization technique where application knowledge regarding data sharing in producer-consumer relationships is used in order to aggressively eliminate unnecessary and predictable snoop-induced cache tag lookups even for references to shared data, thus, achieving significant power reduction with minimal hardware cost. Snoop-induced cache tag lookups for accesses to both shared and private data are eliminated when it is ensured that such lookups will not result in extra knowledge regarding the cache state in respect to the other caches and memories. The proposed methodology relies on the combined support from the compiler, the operating system, and the hardware architecture. Our experiments show average power reductions of more than 80% compared to a general-purpose snoop protocol.

Current communication networks consist of subnetworks of different types. Therefore a common network protocol has to be used for the transmission of data in such a heterogeneous network. Since some time the requirement of mobility in communication networks is showing up. For that reason wireless networks are playing an increasing role as subnetworks. On the other hand there is the need for multiplexed transmission of time-critical and non time-critical (normal) data within a heterogeneous network. In this paper we discuss the problem of multiplexed transmission of time-critical and of non time-critical data over a wireless type subnetwork using a common standardized network protocol. Many of the available wireless subnetworks are of low or medium transmission speed and guarantee a fixed transmission bandwidth at the access point. We describe a mechanism to transmit time-critical data in such a type of subnetwork using a connectionless transport and a connectionless network protocol. The concurrent transmission of non time-critical data using a connection oriented transport and the same connectionless network protocol is assumed to be of lower priority; it is scheduled in a way to fill the remaining capacity, which has not been reserved for the transmission of time-critical data. In our discussion we concentrate on the standardized ISO/OSI protocools CLNP as connectionless network protocol, CLTP as connectionless transport protocol and TP4 as connection oriented transport protocol. We propose a header compression protocol and a fragmentation protocol for use on low bandwidth subnetworks.

The author discusses the nature of the superset of copyback cache protocols described by the MOESI state model, the superset bus transactions of the Futurebus, and the complications encountered in an actual VLSI implementation. It has been shown that the Futurebus provides the best generic and standard solution to cost-effective interconnection of coherent copyback caches. Its fully compelled protocol and the ability to extend or abort transactions allowed the design of the cache controller to be much simpler and less subject to critical timing constraints that it would otherwise have been. The practical design details of the cache controller uncovered numerous coherence 'holes' involving data that is in transit in the fill or flush buffer, or that is partially within the SRAM (static random-access memory) array and partially within the fill buffer. These were solved by thinking of those buffers as one-line caches, with cache state (such as ownership) associated with them. >

Time-dependent groundwater modeling using spreadsheet simulation (TGMSS) model is developed as solution technique. It is a practical method that uses spreadsheets instead of the conventional solution methods. All of the aquifer parameters can easily be described in TGMSS model. The results of TGMSS are validated with MODFLOW. Results showed that TGMSS and MODFLOW results were in good agreement in terms of resulting values of hydraulic heads. © 2005 Wiley Periodicals, Inc. Comput Appl Eng Educ 13: 192–199, 2005; Published online in Wiley InterScience (www.interscience.wiley.com); DOI 10.1002/cae.20048

Replica control protocols in distributed database systems are responsible for the management and maintenance of replicated (redundant) data. Maintaining replicated data improves performance and increases availability. Keeping the replicated data consistent and available during the presence of failures can however become quite difficult. The paper presents the design of a hybrid replica control protocol that attempts to maximise availability and minimise communication overhead, by combining the advantages of two common replica control protocols into one. The protocol was simulated using SimJava, a process based discrete event simulation package. The results from the simulations showed that not only did the hybrid algorithm maintain a high level of availability, it did so while minimising communication overheads.