F. Mattern. Algorithms for distributed termination detection. Distributed Computing. 2,3 (1987).  Home/Search   Document Not in Database   Summary   Related Articles   Check

....first example shows how the observation can be used to express communication patterns, especially, we show how the famous alternating bit protocol can be stated and proved. The second example illustrates the use of superposed observations. We show how a variant of Mattern s termination algorithm [12] can be stated. The interesting aspect of the solution is that we show that communication patterns are reused . The third example shows how the observation mechanism can be used to refine a centralized communication into a distributed one. 4.1 Communication Patterns In this section, we show ....

F. Mattern. Algorithms for distributed termination detection. Distributed Computing, 2:161--175, 1987.

....parallel layer. Finally, Subsection 4.6 describes the proper detection of the end of the search. When there are multiple clusters, detecting determination is a trickier problem than it might rst appear; we adapt some standard techniques from the literature of parallel asynchronous algorithms [24] to detect termination reliably. 4.1. Parallel object oriented design The parallel layer s capabilities are encapsulated in the classes parallelBranching and parallelBranchSub, which have the same function as branching and branchSub, respectively, except that they perform parallel search of the ....

....of optimality, or possibly an incorrect solution. This phenomenon can also occur if there is only a single cluster, but there are application speci c threads that send interprocessor messages. To prevent such premature termination, we use a variant of the four counter method due to Mattern [24], which appears to be the most ecient technique available (the name is misleading, since it is shown in the original reference [24] that the method can be implemented with only three counters) In PICO s case, the procedure works as follows: at the end of the survey sweep, the load balancer ....

[Article contains additional citation context not shown here]

F. Mattern, Algorithms for distributed termination detection, Distributed Computing 2 (1987) 161-175.

....process and the messages in each channel. Recording the global state of a distributed system is an important problem and it finds applications in several aspects of distributed system design. Some such applications include the detection of stable properties such as deadlock [9] and termination [12]; checkpoint recovery from failures [8] debugging distributed software, by resetting the system state to a consistent global state and restarting execution from that state [5, 6] and transformation of an algorithm for solving a static network problem into one that solves the dynamic version of ....

F. Mattern. Algorithms for distributed termination detection. Dist. Comp., 2(3):161--175, 1987.

....specialization process. These communications are asynchronous i.e. the sendings and the receptions of sets of states are non blocking. We use an implementation of MPI [7] for the communications between processes and the termination of the computations is detected using a standard algorithm (see [6]) Currently our prototype assigns statically a partition (a set of locations of the specialized model) to each process. Using this implementation, we obtain preliminary results for two values of n (the number of local iterations between the communications) In our previous works, the memory ....

F. Mattern. Algorithms for Distributed Termination Detection. Distributed Computing,2,1987.

....the next deadlock. A central controller is assumed in their algorithm, thus violating a distributed computing principle. To avoid a single resource (controller) to become a communication performance bottleneck during deadlock detection, any general distributed termination detection algorithm [50] or distributed deadlock detection algorithm [17] could be used instead. In an algorithm described by Misra [52] a special message called marker circulates through GLP to detect and correct deadlock. A cyclic path for traversing all ch i;j 2 CH is precomputed and LPs are initially colored white. ....

F. Mattern. Algorithms for distributed termination detection. Distributed Computing, 2:161--175, 1987.

....or explicitly control messages, might be sent by the algorithm. Many termination detection algorithms have been proposed (e.g. for diffusion computations [5, 16] for synchronous communications [6, 8] for asynchronous but FIFO communications [2, 17] for asynchronous NON FIFO communications [9, 10, 11, 12, 14, 15], and for dynamic systems [4, 11] The worst case message complexity for this problem is f(M) if the underlying system is started from one process) 3] f(M d n) if the underlying system can be started from any number of processes) 12] or f(M E) if the detection algorithm is launched later ....

F. Mattern, "Algorithms for Distributed Termination Detection," Distributed Computing 2, Nov. 1987, pp. 161-175.

....predicates (i.e. the monotone facts ) are a subset of these predicates. Many (perhaps too many ) distributed algorithms to detect such stable predicates (i.e. to decide whether the predicate already holds) have been reported in the literature. A prominent example is termination detection [3], for which a surprising variety of algorithms with various characteristics have been published in recent years. A distributed computation is said to be terminated when all processes are passive and no message is in transit. However, since a passive process may be reactivated when a message is ....

Mattern, F., Algorithms for Distributed Termination Detection. Distributed Computing 2, 1987, pp. 161-175

....an incoming request from PE j then (I; I ) split(I) send I to PE j Figure 1: Basic algorithm for asynchronous random polling. Concurrently, a distributed termination detection protocol is run which recognizes when all PEs have run out of work. We have adapted the four counter method [20] for this purpose. Each PE counts the number of sent and received messages which contain nonempty subproblems. When the global sum over these two counts yields identical results over two global addition rounds, there cannot be any work left (not even in transit) Instead of the ring based summing ....

....Each PE counts the number of sent and received messages which contain nonempty subproblems. When the global sum over these two counts yields identical results over two global addition rounds, there cannot be any work left (not even in transit) Instead of the ring based summing scheme proposed in [20], we use a tree based asynchronous global reduction operation. This is a simple and portable way to bound the termination detection delay by O(T rout log P ) We do not explicitly handle reporting results of the computation here since this is quite cheap for many applications. For example, for ....

F. Mattern. Algorithms for distributed termination detection. Distributed Computing, 2:161--175, 1987.

....answered with an empty subproblem. Note that an empty subproblem can be coded by a short message equivalent to a rejection of the request. Concurrently, a distributed termination detection protocol is run that recognizes when all PEs have run out of work. We have adapted the four counter method [22] for this purpose. Each PE counts the number of sent and received messages that contain nonempty subproblems. When the global sum over these two counts yields identical results over two global addition rounds, there cannot be any work left (not even in transit) Instead of the ring based summing ....

....Each PE counts the number of sent and received messages that contain nonempty subproblems. When the global sum over these two counts yields identical results over two global addition rounds, there cannot be any work left (not even in transit) Instead of the ring based summing scheme proposed in [22], we use a tree based asynchronous global reduction operation. This is a simple and portable way to bound the termination detection delay by O(T rout log P ) If the machine supports it, explicit polling can be replaced by more efficient and more elegant interrupt mechanisms which (almost) only ....

F. Mattern. Algorithms for distributed termination detection. Distributed Computing, 2:161--175, 1987.

....to become an integral part of the distributed system. Note that the presence of boot mirrors does not obsolete the boot server: one of the boot mirrors is considered to be the boot server. Its role is to launch the global wavebased termination detection algorithm similar to the one found in [7] and refined in [2] While a single boot server can take all the decisions such as assigning a Partition ID 3 for the new partition by itself, the various boot mirrors have to negotiate to avoid possible race conditions. One potential problem is if two instances of the same partition try to ....

F. Mattern, "Algorithms for distributed termination detection," Distributed Computing, vol. 2, no. 3, pp. 161--175, 1987.

....of f 54 M;P ) where M is the number of primary messages used by the primary computation and D is the diameter of the target topology. 1. 2 Previous Work and New Results A number of termination detection algorithms with di erent performance characteristics have been proposed in previous work [2, 10, 11, 12, 14, 16]. Some of these algorithms are ecient in terms of the number of secondary messages used [2, 14] but may take a long time to detect termination [2] or incur signi cant computational overhead [14] Similarly, some of the algorithms are either less compute intensive [12] or require less memory [11] ....

F. Mattern, \Algorithm for Distributed Termination Detection," Distributed Computing , Vol. 2, pp. 161-175, 1987.

.... terminating a distributed application is not a trivial issue [7, 6] To summarize the problem, an application can be globally terminated only when all the partitions are locally ready to terminate and there is no message in transit on the network that can potentially wake up one of the partitions [17]. 79 However, an Ada distributed application can be composed of a dynamic number of partitions; when using a client server model, the number of partitions that will compose the distributed program is not known in advance. A server cannot guess whether a new client is going to connect or not. The ....

F. Mattern. Algorithms for distributed termination detection. Distributed Computing, 2(3):161--175, 1987.

....(6) the entry is both written into the transposition table and into the local job queue (7) If the state is already in the table (8) the state is a transposition, and need not be searched again. A global termination detection algorithm (not shown) is used to detect the end of an IDA iteration [9]. In its original formulation, IDA backpropagates search results. These search results can be written into a transposition table, after a state is searched. However, many applications also perform well without regarding the search results. Our TDS implementation does not backpropagate search ....

F. Mattern. Algorithms for Distributed Termination Detection. Distributed Computing, 2:161--175, 1987.

....the list L i ; otherwise, m is related to termination detection (see below) Thus, the Bcg le B i will contain all Lts transitions whose target states are explored by machine i. In order to detect the termination of the parallel Lts generation, we use a virtual ring based algorithm inspired by [Mat87] According to the general denition, global) termination is reached when all local computations are nished (i.e. each machine i has neither remaining states to explore, nor transitions to write in its Bcg le B i ) and all communication channels are empty (i.e. all sent transitions have been ....

F. Mattern. Algorithms for Distributed Termination Detection. Distributed Computing, 2:161175, 1987.

....some way to inspect the state of the ORB s internal threads of control. Example GLADE provides a global termination algorithm for DSA. If all nodes have terminated their own work, and no message is in transit in the communication network, it can be proven that no new computing work can happen [9]. In that case, global termination can be decided. Global or local termination policies can be selected per partition [15] Interface Repository Definition An application may need to invoke operations on objects whose interface was not known when the application was created. For this invocation ....

F. Mattern. Algorithms for Distributed Termination Detection. Distributed Computing, 2(3):161--175, 1987.

....a server to know, or multicast messages to, all other servers in the system. 6 Termination detection: Knowing that the global auction has has not terminated is equivalent to solving termination detection problem in a distributed setting which is not a trivial task though algorithms do exist [18, 19] but are typically message expensive. Hence, the system structuring we decide on, must assist solving this problem in economical ways. Market Shrinkage: Imagine that, in a particular local market, there is no bidding at all right from the start of the auction, or bidding ceases early during the ....

F Mattern, "Algorithms for Distributed Termination Detection", Distributed Computing, Vol 2 (3), 1987, pp. 161-175.

....the list L i # otherwise, m is related to termination detection (see below) Thus, the Bcg file B i will contain all Lts transitions whose target states are explored bymachine i. In order to detect the termination of the parallel Lts generation, we use a virtual ring based algorithm inspired by [29]. According to the general definition, global) termination is reached when all local computations are finished (i.e. each machine i has neither remaining states to explore, nor transitions to write in its Bcg file B i ) and all communication channels are empty (i.e. all sent transitions have ....

F. Mattern. Algorithms for Distributed Termination Detection. Distributed Computing, 2:161--175, 1987.

....within the framework of the UNITY formalism and the assistance of the PVS proof system. The correctness of the algorithm is established through a refinement of an abstract model. 1 Introduction Termination detection of diffusing computations has been extensively studied [13] Many algorithms [5, 9, 10] have been published to deal with this problem. We propose an algorithm which is based on the observation of final paths of a tree where the nodes represent the local computations, and the edges represent message communication. The main features of the algorithm are: no assumption about message ....

Friedemann Mattern. Algorithms for distributed termination detection. Distributed Computing, 2(3):161--175, 1987.

....The correctness of the algorithm is established through a refinement of an abstract model. Inherited properties from this abstract model allow us to derive those of the algorithm. 1 Introduction Termination detection of diffusing computations has been extensively studied [13] Many algorithms [5, 9, 10] have been published to deal with this problem. In general, we can distinguish two types of algorithms: those that are based on counting the exchanged messages, and those that are based on the observation of a distributed dynamic data structure underlying the overall computation. We propose an ....

Friedemann Mattern. Algorithms for distributed termination detection. Distributed Computing, 2(3):161--175, 1987.

.... termination detection is a prototype problem from the area of distributed computing, which is closely related to other important problems of the field, such as distributed garbage collection [33] and snapshot computation [6] It has gained considerable interest in recent years, see for example [5, 11, 12, 20, 21, 32]. This section introduces the termination detection problem and demonstrates that straightforward solutions to it can be based on simple ideas whose correctness can be understood intuitively. 2.1 The Problem Consider a distributed system that consists of processes P 1 ; Delta Delta Delta ; P ....

....of consistent cut and hence of stable property for telepathic computations. For example, the algorithm of Dijkstra et al. 11] requires some control activity whenever a process sends a message, namely coloring the sending process black. In other schemes, activation messages must be counted [20] or acknowledged [12] 7 to be active already at the moment the message is sent (instead of being activated when the message is actually received) Since by rule R1 the sender is also active at the moment of sending the message, rule R2 is observed. Asynchronous Communication. The sticky ....

[Article contains additional citation context not shown here]

Mattern, F. Algorithms for distributed termination detection. Distributed Computing 2 (1987), 161--175.

No context found.

F. Mattern. Algorithms for distributed termination detection. Distributed Computing. 2,3 (1987).

No context found.

Mattern, F.: Algorithms for Distributed Termination Detection. Distributed Computing (DC) 2 (1987) 161-175

No context found.

Friedmann Mattern. Algorithms for distributed termination detection. Distributed Computing, 2:161--175, 1987.

No context found.

F. Mattern. Algorithms for distributed termination detection. Distributed Computing, (2):161--175, 87.

No context found.

F. Mattern. Algorithms for distributed termination detection. Distributed Computing, (2):161--175, 87.

First 50 documents  Next 50

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