#### DMCA

## Performance Model Checking Scenario-Aware Dataflow. (2011)

Venue: | FORMATS, LNCS 6919, |

Citations: | 2 - 0 self |

### Citations

998 |
The semantics of a simple language for parallel programming
- Kahn
- 1974
(Show Context)
Citation Context ...al properties such as absence of deadlock [12] and the key performance properties of throughput [38], [13], latency [14] and buffer occupancy [31]. However, SDF only allows for expressing a fixed behavioural pattern. Using SDF therefore requires abstracting from any dynamism in applications such as decoding different types of video frames in MPEG-4, which imply large variations in resource usage. Neglecting such resource usage variations during design-time analysis may lead to substantially overdimensioning [32]. Other model-based design approaches (e.g., [16], [26]) use Kahn Process Networks [20] to capture dynamism. However, relevant properties, such as the minimal storage space needed to avoid deadlock, throughput and latency, cannot be determined at design-time [25], [10]. The formalism of Scenario-Aware Dataflow (SADF) [35] has recently been introduced as an alternative that does allow modelling dynamic behavior, while various design-time analysis techniques are available and efficient implementations can be created [33]. Current state-of-the-art techniques for SADF have been implemented in the SDF3 tool suite [30], [34]. It supports efficient techniques for worst-case performance... |

596 | D.G.: Static scheduling of synchronous data flow programs for digital signal processing
- Lee, Messerschmitt
- 1987
(Show Context)
Citation Context ...rence in the underlying time models of SADF and IMC. Application of our approach to an MPEG-4 decoder revealed state space reduction factors up to about 21 but also some limitations in terms of scalability and the performance properties that could be analysed. I. Introduction Model-based design of distributed (multi-core) realisations of streaming applications like MPEG-4 and MP3 decoding often use dataflow-based analysis and synthesis techniques [3], [16], [22], [23], [26], [32]. Many such approaches (e.g., [3], [22], [23]) rely on the formalism of (Homogeneous) Synchronous Dataflow ((H)SDF) [21], [29], which is also known as (Weighted) Marked Graphs in Petri Net theory. A wide range of design-time analysis techniques have been developed for SDF, covering both functional properties such as absence of deadlock [12] and the key performance properties of throughput [38], [13], latency [14] and buffer occupancy [31]. However, SDF only allows for expressing a fixed behavioural pattern. Using SDF therefore requires abstracting from any dynamism in applications such as decoding different types of video frames in MPEG-4, which imply large variations in resource usage. Neglecting such resource... |

463 | Introduction to the ISO specification language LOTOS
- Bolognesi
(Show Context)
Citation Context ...37]. Its effect is illustrated in the lower part of Fig. 4. The next section discusses constructing the IMCbased state space of the MPEG-4 example with CADP. V. Formal Verification with CADP The CADP tool suite provides ample means for the three phases of traditional model checking approaches: state-space generation, state-space reduction and (qualitative and quantitative) analysis. The main benefit of CADP is its ability to exploit compositionality. State spaces can be obtained and be reduced in a component-based manner. All this relies on expressing behaviour in an extended version of LOTOS [2]. CADP supports action transitions, probabilistic transitions and Markovian transitions (but not the time transitions as defined in SADF). Given the natural match with two types of these transitions, IMCs are easily expressible without additional semantic modifications [6]. a) State-Space Generation.: The main steps of mapping SADF models onto the analysis capabilities of CADP are provided in Fig. 5. First, the IMC models of each individual SADF actor are glued together by means of parallel composition. The action transitions allow a flexible way of combining IMCs such that channels are for in... |

221 |
Embedded Multiprocessors: Scheduling and synchronization
- Sriram, Bhattacharyya
(Show Context)
Citation Context ...in the underlying time models of SADF and IMC. Application of our approach to an MPEG-4 decoder revealed state space reduction factors up to about 21 but also some limitations in terms of scalability and the performance properties that could be analysed. I. Introduction Model-based design of distributed (multi-core) realisations of streaming applications like MPEG-4 and MP3 decoding often use dataflow-based analysis and synthesis techniques [3], [16], [22], [23], [26], [32]. Many such approaches (e.g., [3], [22], [23]) rely on the formalism of (Homogeneous) Synchronous Dataflow ((H)SDF) [21], [29], which is also known as (Weighted) Marked Graphs in Petri Net theory. A wide range of design-time analysis techniques have been developed for SDF, covering both functional properties such as absence of deadlock [12] and the key performance properties of throughput [38], [13], latency [14] and buffer occupancy [31]. However, SDF only allows for expressing a fixed behavioural pattern. Using SDF therefore requires abstracting from any dynamism in applications such as decoding different types of video frames in MPEG-4, which imply large variations in resource usage. Neglecting such resource usage... |

136 | Bounded Scheduling of Process Networks
- Parks
- 1995
(Show Context)
Citation Context ... expressing a fixed behavioural pattern. Using SDF therefore requires abstracting from any dynamism in applications such as decoding different types of video frames in MPEG-4, which imply large variations in resource usage. Neglecting such resource usage variations during design-time analysis may lead to substantially overdimensioning [32]. Other model-based design approaches (e.g., [16], [26]) use Kahn Process Networks [20] to capture dynamism. However, relevant properties, such as the minimal storage space needed to avoid deadlock, throughput and latency, cannot be determined at design-time [25], [10]. The formalism of Scenario-Aware Dataflow (SADF) [35] has recently been introduced as an alternative that does allow modelling dynamic behavior, while various design-time analysis techniques are available and efficient implementations can be created [33]. Current state-of-the-art techniques for SADF have been implemented in the SDF3 tool suite [30], [34]. It supports efficient techniques for worst-case performance analysis of a relevant subset of SADF based on a max-plus algebraic semantics [11] as well as specialised exact model-checking based techniques for any SADF model to evaluate ... |

120 |
Interactive Markov Chains: The Quest for Quantified Quality,
- Hermanns
- 2002
(Show Context)
Citation Context ...acking flexible support for evaluating user-defined properties (i.e., properties other than certain predefined ones). This paper reports on an investigation to apply the Construction and Analysis of Distributed Processes (CADP) tool suite [9] for quantitative analysis and qualitative model checking of SADF. This enables the usage of powerful (compositional) state-space reduction techniques as well as checking a large range of functional properties expressed in the temporal logic XTL. To enable using CADP, we introduce a compositional semantics for SADF based on Interactive Markov Chains (IMC) [17], [18], thereby formalising previously noticed similarities between two actor types of SADF as well as revealing a fundamental difficulty in capturing the time model of SADF. We propose solutions to handle probabilistic choice and potentially unbounded buffers in CADP. Application of CADP’s model checking capabilities to an MPEG-4 decoder [35] shows that considering dynamism easily yields a state-space explosion, while also substantial state-space reductions can be achieved. Although the variety of analysable properties is enlarged, determining throughput and buffer occupancy suffers from the ... |

46 | Throughput analysis of synchronous data flow graphs.
- Ghamarian
- 2006
(Show Context)
Citation Context ...-based design of distributed (multi-core) realisations of streaming applications like MPEG-4 and MP3 decoding often use dataflow-based analysis and synthesis techniques [3], [16], [22], [23], [26], [32]. Many such approaches (e.g., [3], [22], [23]) rely on the formalism of (Homogeneous) Synchronous Dataflow ((H)SDF) [21], [29], which is also known as (Weighted) Marked Graphs in Petri Net theory. A wide range of design-time analysis techniques have been developed for SDF, covering both functional properties such as absence of deadlock [12] and the key performance properties of throughput [38], [13], latency [14] and buffer occupancy [31]. However, SDF only allows for expressing a fixed behavioural pattern. Using SDF therefore requires abstracting from any dynamism in applications such as decoding different types of video frames in MPEG-4, which imply large variations in resource usage. Neglecting such resource usage variations during design-time analysis may lead to substantially overdimensioning [32]. Other model-based design approaches (e.g., [16], [26]) use Kahn Process Networks [20] to capture dynamism. However, relevant properties, such as the minimal storage space needed to avoid ... |

42 | CADP 2010: A Toolbox for the Construction and Analysis of Distributed Processes.
- Garavel
- 2011
(Show Context)
Citation Context ...SADF based on a max-plus algebraic semantics [11] as well as specialised exact model-checking based techniques for any SADF model to evaluate predefined performance properties like throughput, expected response time and maximum buffer occupancy [37], [34]. The latter type of analysis does however not use existing (quantitative) model checking tools, thereby lacking flexible support for evaluating user-defined properties (i.e., properties other than certain predefined ones). This paper reports on an investigation to apply the Construction and Analysis of Distributed Processes (CADP) tool suite [9] for quantitative analysis and qualitative model checking of SADF. This enables the usage of powerful (compositional) state-space reduction techniques as well as checking a large range of functional properties expressed in the temporal logic XTL. To enable using CADP, we introduce a compositional semantics for SADF based on Interactive Markov Chains (IMC) [17], [18], thereby formalising previously noticed similarities between two actor types of SADF as well as revealing a fundamental difficulty in capturing the time model of SADF. We propose solutions to handle probabilistic choice and potenti... |

37 | SVL: A Scripting Language for Compositional Verification.
- Garavel, Lang
- 2001
(Show Context)
Citation Context ...is reduced using branching bisimulation [15]. In case no non-determinism is left, i.e., no state has two (or more) outgoing τ -transitions —this holds for any SADF model thanks to action-determinism [35]— the resulting IMC is a stochastic process that after amalgamating probabilistic with subsequent Markovian transitions [28] (illustrated in Fig. 6) yields a continuous-time Markov chain (CTMC). CADP allows qualitative model checking on any IMC, including the non-deterministic ones; quantitative analysis however is restricted to CTMCs. All above steps are realised by the scripting language SVL [8] of CADP. b) State-Space Reductions.: As the SADF buffers are modeled by data types and are in principle unbounded, they do not have a fixed a priori capacity. This means that state-space generation may not terminate. Specific characteristics of the MPEG-4 example (which include consistency between production and consumption rates and the cyclic dependencies) imply a bounded system, but indications about the size of the individual buffers are not available by a static analysis of the SADF model. In case of bounded SADF systems, CADP can generate the state space, possibly apply some reductions,... |

36 | Throughput-buffering trade-off exploration for cyclo-static and synchronous dataflow graphs.
- Stuijk, Geilen, et al.
- 2008
(Show Context)
Citation Context ...) realisations of streaming applications like MPEG-4 and MP3 decoding often use dataflow-based analysis and synthesis techniques [3], [16], [22], [23], [26], [32]. Many such approaches (e.g., [3], [22], [23]) rely on the formalism of (Homogeneous) Synchronous Dataflow ((H)SDF) [21], [29], which is also known as (Weighted) Marked Graphs in Petri Net theory. A wide range of design-time analysis techniques have been developed for SDF, covering both functional properties such as absence of deadlock [12] and the key performance properties of throughput [38], [13], latency [14] and buffer occupancy [31]. However, SDF only allows for expressing a fixed behavioural pattern. Using SDF therefore requires abstracting from any dynamism in applications such as decoding different types of video frames in MPEG-4, which imply large variations in resource usage. Neglecting such resource usage variations during design-time analysis may lead to substantially overdimensioning [32]. Other model-based design approaches (e.g., [16], [26]) use Kahn Process Networks [20] to capture dynamism. However, relevant properties, such as the minimal storage space needed to avoid deadlock, throughput and latency, cannot... |

34 | A scenario-aware data flow model for combined long-run average and worst-case performance analysis. In:
- BD, Geilen, et al.
- 2006
(Show Context)
Citation Context ... requires abstracting from any dynamism in applications such as decoding different types of video frames in MPEG-4, which imply large variations in resource usage. Neglecting such resource usage variations during design-time analysis may lead to substantially overdimensioning [32]. Other model-based design approaches (e.g., [16], [26]) use Kahn Process Networks [20] to capture dynamism. However, relevant properties, such as the minimal storage space needed to avoid deadlock, throughput and latency, cannot be determined at design-time [25], [10]. The formalism of Scenario-Aware Dataflow (SADF) [35] has recently been introduced as an alternative that does allow modelling dynamic behavior, while various design-time analysis techniques are available and efficient implementations can be created [33]. Current state-of-the-art techniques for SADF have been implemented in the SDF3 tool suite [30], [34]. It supports efficient techniques for worst-case performance analysis of a relevant subset of SADF based on a max-plus algebraic semantics [11] as well as specialised exact model-checking based techniques for any SADF model to evaluate predefined performance properties like throughput, expected ... |

21 | The Artemis Workbench for System-level Performance Evaluation of Embedded Systems.
- Pimentel
- 2008
(Show Context)
Citation Context ...obabilistic choice and potentially unbounded buffers, both of which are not natively supported, as well as a fundamental difference in the underlying time models of SADF and IMC. Application of our approach to an MPEG-4 decoder revealed state space reduction factors up to about 21 but also some limitations in terms of scalability and the performance properties that could be analysed. I. Introduction Model-based design of distributed (multi-core) realisations of streaming applications like MPEG-4 and MP3 decoding often use dataflow-based analysis and synthesis techniques [3], [16], [22], [23], [26], [32]. Many such approaches (e.g., [3], [22], [23]) rely on the formalism of (Homogeneous) Synchronous Dataflow ((H)SDF) [21], [29], which is also known as (Weighted) Marked Graphs in Petri Net theory. A wide range of design-time analysis techniques have been developed for SDF, covering both functional properties such as absence of deadlock [12] and the key performance properties of throughput [38], [13], latency [14] and buffer occupancy [31]. However, SDF only allows for expressing a fixed behavioural pattern. Using SDF therefore requires abstracting from any dynamism in applications such a... |

20 |
Compositional, and Extensible Framework for Dynamic Fault Tree Analysis.
- Boudali
- 2009
(Show Context)
Citation Context ...[17]. IMCs extend traditional continuoustime Markov chains with the ability to interact using interactive actions. Hence, an IMC is a state-based model with action and Markovian transitions. The Markovian transitions model advancing time according to a negative exponential distribution whereas actions can be used to synchronise with other IMCs in a handshake manner like in CSP [19]. By parallel composition of IMCs, a complete system can be described in a compositional way [17], [18]. IMCs have shown their practical relevance in applications of various domains, ranging from dynamic fault trees [4], standardised engineering languages such as AADL (Architectural Analysis and Design Language) [5] and Statemate [1] to Globally Asynchronous Locally Synchronous (GALS) hardware designs [7]. We exploit IMCs to define the semantics of SADF in a componentbased manner, i.e., each individual actor yields one (or more) IMC(s) which are subsequently put in parallel. Capturing the behaviour of an SADF model such as the MPEG-4 example in Fig. 1 with IMC is not trivial due to a fundamental difference in their underlying time models. The original semantics of SADF is defined in terms of so-called Timed ... |

17 | An Efficient and Complete Approach for Throughput Maximal SDF Allocation and Scheduling on MultiCore Platforms.
- Bonfietti
- 2010
(Show Context)
Citation Context ...ncluded dealing with probabilistic choice and potentially unbounded buffers, both of which are not natively supported, as well as a fundamental difference in the underlying time models of SADF and IMC. Application of our approach to an MPEG-4 decoder revealed state space reduction factors up to about 21 but also some limitations in terms of scalability and the performance properties that could be analysed. I. Introduction Model-based design of distributed (multi-core) realisations of streaming applications like MPEG-4 and MP3 decoding often use dataflow-based analysis and synthesis techniques [3], [16], [22], [23], [26], [32]. Many such approaches (e.g., [3], [22], [23]) rely on the formalism of (Homogeneous) Synchronous Dataflow ((H)SDF) [21], [29], which is also known as (Weighted) Marked Graphs in Petri Net theory. A wide range of design-time analysis techniques have been developed for SDF, covering both functional properties such as absence of deadlock [12] and the key performance properties of throughput [38], [13], latency [14] and buffer occupancy [31]. However, SDF only allows for expressing a fixed behavioural pattern. Using SDF therefore requires abstracting from any dynamis... |

16 | Worst-case Performance Analysis of Synchronous Dataflow Scenarios.
- Geilen, Stuijk
- 2010
(Show Context)
Citation Context ... space needed to avoid deadlock, throughput and latency, cannot be determined at design-time [25], [10]. The formalism of Scenario-Aware Dataflow (SADF) [35] has recently been introduced as an alternative that does allow modelling dynamic behavior, while various design-time analysis techniques are available and efficient implementations can be created [33]. Current state-of-the-art techniques for SADF have been implemented in the SDF3 tool suite [30], [34]. It supports efficient techniques for worst-case performance analysis of a relevant subset of SADF based on a max-plus algebraic semantics [11] as well as specialised exact model-checking based techniques for any SADF model to evaluate predefined performance properties like throughput, expected response time and maximum buffer occupancy [37], [34]. The latter type of analysis does however not use existing (quantitative) model checking tools, thereby lacking flexible support for evaluating user-defined properties (i.e., properties other than certain predefined ones). This paper reports on an investigation to apply the Construction and Analysis of Distributed Processes (CADP) tool suite [9] for quantitative analysis and qualitative mod... |

16 |
Multiprocessor resource allocation for hard-real-time streaming with a dynamic jobmix,”
- Moreira, Mol, et al.
- 2005
(Show Context)
Citation Context ...ith probabilistic choice and potentially unbounded buffers, both of which are not natively supported, as well as a fundamental difference in the underlying time models of SADF and IMC. Application of our approach to an MPEG-4 decoder revealed state space reduction factors up to about 21 but also some limitations in terms of scalability and the performance properties that could be analysed. I. Introduction Model-based design of distributed (multi-core) realisations of streaming applications like MPEG-4 and MP3 decoding often use dataflow-based analysis and synthesis techniques [3], [16], [22], [23], [26], [32]. Many such approaches (e.g., [3], [22], [23]) rely on the formalism of (Homogeneous) Synchronous Dataflow ((H)SDF) [21], [29], which is also known as (Weighted) Marked Graphs in Petri Net theory. A wide range of design-time analysis techniques have been developed for SDF, covering both functional properties such as absence of deadlock [12] and the key performance properties of throughput [38], [13], latency [14] and buffer occupancy [31]. However, SDF only allows for expressing a fixed behavioural pattern. Using SDF therefore requires abstracting from any dynamism in applications ... |

14 | A predictable multiprocessor design flow for streaming applications with dynamic behaviour,”
- Stuijk
- 2010
(Show Context)
Citation Context ...istic choice and potentially unbounded buffers, both of which are not natively supported, as well as a fundamental difference in the underlying time models of SADF and IMC. Application of our approach to an MPEG-4 decoder revealed state space reduction factors up to about 21 but also some limitations in terms of scalability and the performance properties that could be analysed. I. Introduction Model-based design of distributed (multi-core) realisations of streaming applications like MPEG-4 and MP3 decoding often use dataflow-based analysis and synthesis techniques [3], [16], [22], [23], [26], [32]. Many such approaches (e.g., [3], [22], [23]) rely on the formalism of (Homogeneous) Synchronous Dataflow ((H)SDF) [21], [29], which is also known as (Weighted) Marked Graphs in Petri Net theory. A wide range of design-time analysis techniques have been developed for SDF, covering both functional properties such as absence of deadlock [12] and the key performance properties of throughput [38], [13], latency [14] and buffer occupancy [31]. However, SDF only allows for expressing a fixed behavioural pattern. Using SDF therefore requires abstracting from any dynamism in applications such as deco... |

13 | Execution-time prediction for dynamic streaming applications with task-level parallelism. In: - Poplavko, Basten, et al. - 2007 |

13 | Probabilistic Branching in Markovian Process Algebras.
- Rettelbach
- 1995
(Show Context)
Citation Context ...Fig. 5. From SADF models to analysis τ -transitions. Thus, as a result either a state has only outgoing τ -transitions or outgoing Markovian transitions. Parts that become unreachable by this operation are removed from the state space. Subsequently, the state space of the IMC is reduced using branching bisimulation [15]. In case no non-determinism is left, i.e., no state has two (or more) outgoing τ -transitions —this holds for any SADF model thanks to action-determinism [35]— the resulting IMC is a stochastic process that after amalgamating probabilistic with subsequent Markovian transitions [28] (illustrated in Fig. 6) yields a continuous-time Markov chain (CTMC). CADP allows qualitative model checking on any IMC, including the non-deterministic ones; quantitative analysis however is restricted to CTMCs. All above steps are realised by the scripting language SVL [8] of CADP. b) State-Space Reductions.: As the SADF buffers are modeled by data types and are in principle unbounded, they do not have a fixed a priori capacity. This means that state-space generation may not terminate. Specific characteristics of the MPEG-4 example (which include consistency between production and consumpti... |

11 | Ten Years of Performance Evaluation for Concurrent Systems using
- Coste
- 2010
(Show Context)
Citation Context ... checking approaches: state-space generation, state-space reduction and (qualitative and quantitative) analysis. The main benefit of CADP is its ability to exploit compositionality. State spaces can be obtained and be reduced in a component-based manner. All this relies on expressing behaviour in an extended version of LOTOS [2]. CADP supports action transitions, probabilistic transitions and Markovian transitions (but not the time transitions as defined in SADF). Given the natural match with two types of these transitions, IMCs are easily expressible without additional semantic modifications [6]. a) State-Space Generation.: The main steps of mapping SADF models onto the analysis capabilities of CADP are provided in Fig. 5. First, the IMC models of each individual SADF actor are glued together by means of parallel composition. The action transitions allow a flexible way of combining IMCs such that channels are for instance shared between a sending and receiving SADF actor. As now the actions to compose the individual IMCs are not of any relevance, they are abstracted away (hiding). This entails that all action transitions are labeled with the internal action τ . Applying the maximal p... |

10 |
Compositional Dependability Evaluation for STATEMATE.
- Bode
- 2009
(Show Context)
Citation Context ...ence, an IMC is a state-based model with action and Markovian transitions. The Markovian transitions model advancing time according to a negative exponential distribution whereas actions can be used to synchronise with other IMCs in a handshake manner like in CSP [19]. By parallel composition of IMCs, a complete system can be described in a compositional way [17], [18]. IMCs have shown their practical relevance in applications of various domains, ranging from dynamic fault trees [4], standardised engineering languages such as AADL (Architectural Analysis and Design Language) [5] and Statemate [1] to Globally Asynchronous Locally Synchronous (GALS) hardware designs [7]. We exploit IMCs to define the semantics of SADF in a componentbased manner, i.e., each individual actor yields one (or more) IMC(s) which are subsequently put in parallel. Capturing the behaviour of an SADF model such as the MPEG-4 example in Fig. 1 with IMC is not trivial due to a fundamental difference in their underlying time models. The original semantics of SADF is defined in terms of so-called Timed Probabilistic Systems [36]. They contain action, probabilistic and deterministic time transitions. The time transiti... |

10 | Latency Minimization for Synchronous Data Flow Graphs.
- Ghamarian
- 2007
(Show Context)
Citation Context ...of distributed (multi-core) realisations of streaming applications like MPEG-4 and MP3 decoding often use dataflow-based analysis and synthesis techniques [3], [16], [22], [23], [26], [32]. Many such approaches (e.g., [3], [22], [23]) rely on the formalism of (Homogeneous) Synchronous Dataflow ((H)SDF) [21], [29], which is also known as (Weighted) Marked Graphs in Petri Net theory. A wide range of design-time analysis techniques have been developed for SDF, covering both functional properties such as absence of deadlock [12] and the key performance properties of throughput [38], [13], latency [14] and buffer occupancy [31]. However, SDF only allows for expressing a fixed behavioural pattern. Using SDF therefore requires abstracting from any dynamism in applications such as decoding different types of video frames in MPEG-4, which imply large variations in resource usage. Neglecting such resource usage variations during design-time analysis may lead to substantially overdimensioning [32]. Other model-based design approaches (e.g., [16], [26]) use Kahn Process Networks [20] to capture dynamism. However, relevant properties, such as the minimal storage space needed to avoid deadlock, thro... |

9 | Efficient SAT-based mapping and scheduling of homogeneous synchronous dataflow graphs for throughput optimization.
- Liu
- 2008
(Show Context)
Citation Context ...ling with probabilistic choice and potentially unbounded buffers, both of which are not natively supported, as well as a fundamental difference in the underlying time models of SADF and IMC. Application of our approach to an MPEG-4 decoder revealed state space reduction factors up to about 21 but also some limitations in terms of scalability and the performance properties that could be analysed. I. Introduction Model-based design of distributed (multi-core) realisations of streaming applications like MPEG-4 and MP3 decoding often use dataflow-based analysis and synthesis techniques [3], [16], [22], [23], [26], [32]. Many such approaches (e.g., [3], [22], [23]) rely on the formalism of (Homogeneous) Synchronous Dataflow ((H)SDF) [21], [29], which is also known as (Weighted) Marked Graphs in Petri Net theory. A wide range of design-time analysis techniques have been developed for SDF, covering both functional properties such as absence of deadlock [12] and the key performance properties of throughput [38], [13], latency [14] and buffer occupancy [31]. However, SDF only allows for expressing a fixed behavioural pattern. Using SDF therefore requires abstracting from any dynamism in applica... |

7 |
Multiprocessor SoC software design flows: a focus on Kahn process networks
- Haid, Huang, et al.
- 2009
(Show Context)
Citation Context ...ed dealing with probabilistic choice and potentially unbounded buffers, both of which are not natively supported, as well as a fundamental difference in the underlying time models of SADF and IMC. Application of our approach to an MPEG-4 decoder revealed state space reduction factors up to about 21 but also some limitations in terms of scalability and the performance properties that could be analysed. I. Introduction Model-based design of distributed (multi-core) realisations of streaming applications like MPEG-4 and MP3 decoding often use dataflow-based analysis and synthesis techniques [3], [16], [22], [23], [26], [32]. Many such approaches (e.g., [3], [22], [23]) rely on the formalism of (Homogeneous) Synchronous Dataflow ((H)SDF) [21], [29], which is also known as (Weighted) Marked Graphs in Petri Net theory. A wide range of design-time analysis techniques have been developed for SDF, covering both functional properties such as absence of deadlock [12] and the key performance properties of throughput [38], [13], latency [14] and buffer occupancy [31]. However, SDF only allows for expressing a fixed behavioural pattern. Using SDF therefore requires abstracting from any dynamism in a... |

6 | A Performance Analysis Tool for Scenario-Aware Streaming Applications.
- Theelen
- 2007
(Show Context)
Citation Context ...sign approaches (e.g., [16], [26]) use Kahn Process Networks [20] to capture dynamism. However, relevant properties, such as the minimal storage space needed to avoid deadlock, throughput and latency, cannot be determined at design-time [25], [10]. The formalism of Scenario-Aware Dataflow (SADF) [35] has recently been introduced as an alternative that does allow modelling dynamic behavior, while various design-time analysis techniques are available and efficient implementations can be created [33]. Current state-of-the-art techniques for SADF have been implemented in the SDF3 tool suite [30], [34]. It supports efficient techniques for worst-case performance analysis of a relevant subset of SADF based on a max-plus algebraic semantics [11] as well as specialised exact model-checking based techniques for any SADF model to evaluate predefined performance properties like throughput, expected response time and maximum buffer occupancy [37], [34]. The latter type of analysis does however not use existing (quantitative) model checking tools, thereby lacking flexible support for evaluating user-defined properties (i.e., properties other than certain predefined ones). This paper reports on an i... |

5 | Quantitative Evaluation in Embedded System Design: Validation of Multiprocessor Multithreaded Architectures.
- Coste
- 2008
(Show Context)
Citation Context .... The Markovian transitions model advancing time according to a negative exponential distribution whereas actions can be used to synchronise with other IMCs in a handshake manner like in CSP [19]. By parallel composition of IMCs, a complete system can be described in a compositional way [17], [18]. IMCs have shown their practical relevance in applications of various domains, ranging from dynamic fault trees [4], standardised engineering languages such as AADL (Architectural Analysis and Design Language) [5] and Statemate [1] to Globally Asynchronous Locally Synchronous (GALS) hardware designs [7]. We exploit IMCs to define the semantics of SADF in a componentbased manner, i.e., each individual actor yields one (or more) IMC(s) which are subsequently put in parallel. Capturing the behaviour of an SADF model such as the MPEG-4 example in Fig. 1 with IMC is not trivial due to a fundamental difference in their underlying time models. The original semantics of SADF is defined in terms of so-called Timed Probabilistic Systems [36]. They contain action, probabilistic and deterministic time transitions. The time transitions denote that time must have advanced with the specified exact amount o... |

3 |
An Overview and
- Nicollin, Sifakis
- 1991
(Show Context)
Citation Context ...ipate in the interaction between IMCs. Therefore, such internal actions are often hidden for the environment. The upper part of Fig. 4 illustrates the effect of applying maximal progress after parallel composition. In Section II, we discussed the concept of self-timed execution for dataflow models as a means to partially resolve non-determinism: an actor is fired without delays in case all its enabled actions are performed prior to advancing time, i.e., not only internal actions that can be hidden for other IMCs. Prioritising any action over progress in time is sometimes called action urgency [24], [37]. Its effect is illustrated in the lower part of Fig. 4. The next section discusses constructing the IMCbased state space of the MPEG-4 example with CADP. V. Formal Verification with CADP The CADP tool suite provides ample means for the three phases of traditional model checking approaches: state-space generation, state-space reduction and (qualitative and quantitative) analysis. The main benefit of CADP is its ability to exploit compositionality. State spaces can be obtained and be reduced in a component-based manner. All this relies on expressing behaviour in an extended version of LOT... |

3 |
Scenario-Aware Dataflow.
- Theelen
- 2008
(Show Context)
Citation Context ...ring languages such as AADL (Architectural Analysis and Design Language) [5] and Statemate [1] to Globally Asynchronous Locally Synchronous (GALS) hardware designs [7]. We exploit IMCs to define the semantics of SADF in a componentbased manner, i.e., each individual actor yields one (or more) IMC(s) which are subsequently put in parallel. Capturing the behaviour of an SADF model such as the MPEG-4 example in Fig. 1 with IMC is not trivial due to a fundamental difference in their underlying time models. The original semantics of SADF is defined in terms of so-called Timed Probabilistic Systems [36]. They contain action, probabilistic and deterministic time transitions. The time transitions denote that time must have advanced with the specified exact amount of time units before the next transitions become enabled. A specific combination of transitions allows expressing any discrete execution time distribution associated to the scenario of an actor, see Fig. 2. The left-hand side shows that after performing some (internal) action τ (think of drawing a sample from a discrete distribution), time advances exactly ti time units with probability pi for i = 1, . . . , n. Notice that such a dist... |

1 |
Dependability and Performance Analysis of Extended
- Bozzano
- 2011
(Show Context)
Citation Context ...ractive actions. Hence, an IMC is a state-based model with action and Markovian transitions. The Markovian transitions model advancing time according to a negative exponential distribution whereas actions can be used to synchronise with other IMCs in a handshake manner like in CSP [19]. By parallel composition of IMCs, a complete system can be described in a compositional way [17], [18]. IMCs have shown their practical relevance in applications of various domains, ranging from dynamic fault trees [4], standardised engineering languages such as AADL (Architectural Analysis and Design Language) [5] and Statemate [1] to Globally Asynchronous Locally Synchronous (GALS) hardware designs [7]. We exploit IMCs to define the semantics of SADF in a componentbased manner, i.e., each individual actor yields one (or more) IMC(s) which are subsequently put in parallel. Capturing the behaviour of an SADF model such as the MPEG-4 example in Fig. 1 with IMC is not trivial due to a fundamental difference in their underlying time models. The original semantics of SADF is defined in terms of so-called Timed Probabilistic Systems [36]. They contain action, probabilistic and deterministic time transitions.... |