S. Poledna. Replica determinism in distributed real-time systems: A brief survey. Real-Time Systems, 6(3):289--316, May 1994.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....replicated nodes can cause different task interleaving. Consequently, different replicas (even if correct) can respond to the same inputs in different orders, providing inconsistent results if inputs are non commutative. That is the problem of Replica Determinism in Distributed Real Time Systems [13]. In the proposed approach, fault tolerance is achieved by node replication, with software based replica management and fault tolerance transparent algorithms. A software layer implemented is the responsible for the transparent incorporation of the fault tolerance mechanisms. The FTManager, is ....

Poledna, S. Replica Determinism in Distributed Real- Time Systems: A Brief Survey. Real-Time Systems, Vol. 6, 3, 1994, 289-316.

....and the CPU cost for processing the information such as updating, retrieving, and storing. Network cost are decomposed into communication and transmission costs. This problem of determining a solution to this problem under varying system conditions has been addressed in a number of publications [ChL87, DoF82, MoL77, Mur83, NoA87, Pol94, Pu86, RoS96, SeS79, Seg76, Yu85 and ZSA92]. Solutions to the file assignment problem can be divided into two classes: static file assignment and dynamic file assignment. In static file assignment [MoL77, Mur83, and Pu86] the design of the file assigner is based upon a set of time invariant system parameters. Therefore, in solutions ....

Poledna S. (1994); "Replica Determinism in Distributed Real-Time Systems: A Brief Survery"; Real Time Systems Vol. 6 No. 3 (1994)

....agreed time base and communication can be synchronized with measurement instants known by all nodes. Furthermore, the Time Triggered Architecture helps in achieving deterministic timing behavior, which is is a key feature for replication of systems or subsystems in order to increase dependability [Pol94]. 2.5 Monitoring Monitoring and debugging of distributed embedded real time systems di#er significantly from debugging and testing programs for desktop computers. While a program written for a desktop computer can display thousands of debugging messages to the programmer using a single (output) ....

S. Poledna. Replica Determinism in Distributed Real-Time Systems: A Brief Survey. RealTime Systems, 6:289--316, 1994.

....error in a program, which stays there until removed. 3.2.1. 3 Failure semantics The above classification of failure modes is not restricted to individual instances of failures, but can be used to classify the failure behavior of components, which is called a component s failure semantics (Poledna [87]) Failure semantics A component exhibits a given failure semantic if the probability of failure modes, which are not covered by the failure semantic, is sufficiently low. If a given component is assumed to have synchronization failure semantics, then all individual failures of the component ....

.... special hardware [113] How to deterministically reproduce the observations using mechanisms in software [22] 63] 104] or mechanisms in hardware [113] The problem of defining a global state in distributed systems [29] using logical clocks [12] 58] or synchronized physical clocks [51] 53][87]. However, the number of references on research regarding monitoring for testing and debugging of single node real time systems, and multiple node (distributed) real time systems, that consistently handle time, distribution, interrupts, clock synchronization, and scheduling, dwindle fast (to ....

Poledna S. Replica Determinism in Distributed Real-Time Systems: A Brief Survey. RealTimes systems Journal, Kluwer A.P., (6):289-316, 1994.

....makes it easier to replicate legacy systems. However, active replication uses processing resources heavily and requires the processing of requests to be deterministic. This last point is a very strong limitation since, in a program, there exist many potential sources of non determinism [27]. For instance, multi threading typically introduces non determinism. A preliminary abstract of this paper appeared in Proceedings of the 17th IEEE International Symposium on Reliable Distributed Systems (IEEE CS Press, pp. 43 50) It is now largely superseded by this version. Determinism ....

....replication techniques presented in Section 4.2. The main motivation for redundant processing is to guarantee a constant response time, regardless of the occurrence of failures. This is hence no surprise that this family of replication techniques is usually favored by real time systems designers [27]. There are two main replication techniques with redundant processing: active replication, and semi active replication. 4.3.1 Active Replication In active replication [22, 30] clients sends their requests to all replicas. All replicas process incoming requests in the same way and in the same ....

S. Poledna. Replica determinism in distributed real-time systems: A brief survey. Real-Time Systems, 6(3):289--316, May 1994.

....scheduling. These factors are often studied in the literature as a problem for the design of fault tolerant systems: by disrupting replica determinism they may cause spurious disagreements between redundant copies of software, and complicate the design of state recovery mechanisms (cf [Poledna 1994]) A system design Bev Littlewood, Lorenzo Strigini: A discussion of practices for enhancing diversity in software designs p. 14 using design or functional diversity will also be affected by unintentional data diversity: for instance, if two diverse redundant channels produce an intermediate ....

S. Poledna, "Replica Determinism in Distributed Real-Time Systems: A Brief Survey", Real-Time Systems Journal, 6, pp. 289-316, 1994.

....are a priori known to all components. All components have access to a global time and communication can be synchronized with measurement instants known by all nodes. Furthermore, deterministic timing behavior is a key feature for replication of systems or subsystems to increase dependability [Pol94] 4 Case Study The time triggered system model introduced in the previous Section was used for the imple4 mentation of a model car, which acts as an autonomous robot with sensory inputs [Pet01] The model comprises a mobile robot ( smart car ) equipped with a suit of pivoted distance sensors, ....

S. Poledna. Replica determinism in distributed real-time systems: A brief survey. Real-Time Systems, 6:289--316, 1994.

....as well. 3 The experiment Active replication Our experiment consists of a replicated server and several clients. Each client repeatedly sends a request to the replicated server and waits for a reply. The server is replicated by means of active replication (also called state machine approach) [Sch90, Sch93, Pol94]. In active replication, clients use atomic broadcast to send their requests to the replicas. Atomic broadcast ensures that all server replicas receive the client requests in the same order. Upon reception of a request, each server replica performs the same deterministic processing (in our case, ....

S. Poledna. Replica determinism in distributed real-time systems: A brief survey. Real-Time Systems, 6(3):289--316, May 1994.

....form of force summing. For example, the outputs of different channels may energize separate coils of a single solenoid, or multiple hydraulic pistons may be linked to a single shaft. There are three key elements to this architecture; collectively they ensure and exploit replica determinism [19]. 1. Clock synchronization ensures that all nonfaulty channels can coordinate their actions and communications using a common notion of time. Synchronization between nonfaulty channels must be achieved in the presence of disruption caused by faulty channels. The number of faulty channels that can ....

Stefan Poledna. Replica determinism in distributed real-time systems: A brief survey. Real-Time Systems, 6:289--316, 1994.

....replicated nodes can cause different task interleaving. Consequently, different replicas (even if correct) can respond to the same inputs in different orders, providing inconsistent results if inputs are non commutative. That is the problem of Replica Determinism in Distributed Real Time Systems [13]. In the proposed approach, fault tolerance is achieved by node replication, with software based replica management and fault tolerance transparent algorithms. A software layer implemented between the application and the real time kernel, the Fault Tolerance Manager (FTManager) Replica ....

Poledna, S. Replica Determinism in Distributed Real-Time Systems: A Brief Survey. Real-Time Systems, Vol. 6, 3, 1994, 289-316.

....caused by an error in a program, which stays there until removed. 3. 3 Failure semantics The above classification of failure modes is not restricted to individual instances of failures, but can be used to classify the failure behavior of components, which is called a component s failure semantics [48]. Failure semantics A component exhibits a given failure semantic if the probability of failure modes, which are not covered by the failure semantic, is sufficiently low. 10 If a given component is assumed to have synchronization failure semantics, then all individual failures of the ....

.... using special hardware [68] How to deterministically reproduce the observations using mechanisms in software [12] 35] 61] or mechanisms in hardware [68] The problem of defining a global state in distributed systems [16] using logical clocks [7] 32] or synchronized physical clocks [26] 28][48]. However, the number of references on research regarding monitoring for testing and debugging of single node real time systems, and multiple node (distributed) real time systems, that consistently handle time, distribution, interrupts, clock synchronization, and scheduling, dwindle fast (to ....

Poledna S. Replica Determinism in Distributed Real-Time Systems: A Brief Survey. RealTimes systems Journal, Kluwer A.P., (6):289-316, 1994.

No context found.

S. Poledna. Replica determinism in distributed real-time systems: A brief survey. Real-Time Systems, 6(3):289--316, May 1994.

No context found.

S. Poledna. Replica determinism in distributed real-time systems: A brief survey. Real-Time Systems, 6(3):289--316, May 1994.

No context found.

POLEDNA, S. 1994. Replica determinism in distributed real-time systems: A brief survey. RealTime Systems 6, 3 (May), 289--316.

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