P.M. Merlin and P.J. Schweitzer. Deadlock avoidance in store-and-forward networks: Store and forward deadlock. IEEE Trans. on Commun., COM-28:345--352, March 1980.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....network resources. In this note we concentrate on one approach to deadlock prevention that has been extensively studied in the literature. This approach involves solutions based on dividing the buffer pool into buffer classes and utilizing these classes so as to prevent cyclic waiting chains [Gop84, MS80, TU81, BC89]. Some of these solutions are based on restricting the family of allowed routes in order to avoid deadlocks. The problem addressed in this note involves two basic characteristics of deadlock prevention policies, namely, their buffer requirements and the quality of the allowed routes. In order to ....

....buffer requirements and the quality of the allowed routes. In order to present the problem, let us compare these characteristics as manifested in routing schemes founded on two opposing philosophies for buffer allocation proposed in the literature. The first philosophy, governing the schemes of [Gun81, Gop84, MS80], features complete independence of the network topology (except for knowing the network diameter D) Generally speaking, schemes based on this philosophy do not interfere with the routing itself, and thus allow sending packets along shortest paths, for instance. However, at least cn buffers per ....

[Article contains additional citation context not shown here]

P. M. Merlin and P. J. Schweitzer. Deadlock avoidance in store-and-forward networks--I: Store and forward deadlock. IEEE Trans. on Commun., COM-28:345--352, March 1980.

....of possible deadlocks. In this paper we show that there is an algorithm that takes an SCI configuration as input, and decides whether or not this configuration is prone to deadlocks. The algorithm is derived from corresponding ones from store and forward networks, and wormhole routing networks [Gel81, Gn81, MS80, DS87]. Then we present an algorithm that alters the routing functions of a configuration that is prone to deadlocks, in order to make it deadlock free. Our results indicates the following approach for configuring and reconfiguring of SCI installations: First create your configuration while considering ....

P. M. Merlin and P. J. Schweitzer. Deadlock avoidance in store-and-forward networks -- i: Store-and-forward deadlock. IEEE Transactions on Communications, COM-28:345--354, March 1980.

....an increase in the hardware complexity of the node. B. Deadlock Free Routing Using Virtual Channels A well known approach in store and forward packet networks to avoid deadlocks consists of defining a partial order on the buffers (hierarchy of packet buffer pools) at each node [7] 9] 10] [17], 28] 29] and allocating them on various criteria like number of hops traveled, the route which the packet is travelling, etc. An approach similar to the hierarchy of buffer pools, called the virtual channels technique [5] combines restricted routing and buffer partition to achieve ....

P. M. Merlin and P. J. Schweitzer, "Deadlock avoidance in store-and forward networks---I: Store-and-forward deadlock," IEEE Trans. Commun., vol. COM-28, pp. 345--354, Mar. 1980.

....influence not only the efficiency of the routing strategy but also its correctness. Several techniques have been developed to design deadlock free routing strategies in which deadlocks are avoided by ordering the buffers and allowing each message to use them in a monotonically increasing fashion ([12, 11, 14, 5, 1, 8, 7, 13, 2, 3, 4, 9] among the others) This idea results in the generation of a directed acyclic resource dependencies graph (DAG) thus preventing deadlock configurations. A DAG based method has been introduced in [11, 15] and furtherly studied in [6] in which the ordering in the set of buffers of each vertex is ....

P.M. Merlin and P.J. Schweitzer. Deadlock avoidance in store-and-forward networks: Store and forward deadlock. IEEE Trans. on Commun., COM-28:345--352, March 1980.

....and achieves O(n) communication complexity if the network stabilizes. 1.1 Related work Techniques The algorithms in this paper combine several techniques from recently reported research. The slide mechanism combines in an interesting way ideas from [AMS89] with a technique appearing in [MS80] Combining slide with a concept introduced in [AFWZ88, AAF ] we construct an end to end algorithm, the majority algorithm. This algorithm, in conjunction with the information dispersal algorithm (IDA) of Rabin [Rab89] yields our data dispersal algorithm. In the second part of the paper we ....

P. M. Merlin and P. J. Schweitzer. Deadlock avoidance in store-and-forward networks 1: Store-and-forward deadlock. IEEE Transaction on Communications, 28(3):345--354, March 1980.

....to outgoing links. This new network structure requires careful re examination of classical basic tasks in communication networks, e.g. CG88, CGK88, AC 94, DK 95] Deadlock prevention and lossless networks: Deadlock prevention is one of the most important issues in distributed computing, e.g. [MS80, TU81, Ci89, MC82]. Store and forward deadlock is a situation in which there is a cycle of processors such that each processor is full with packets and has to transmit its packets to the next processor in the cycle. In such deadlock situation no processor can transmit or receive packets, therefore no progress is ....

P. M. Merlin and P. J. Schweitzer, "Deadlock avoidance in store-and-forward networks --- I: Store and forward deadlock," IEEE Transactions on Communication, vol. COM-28, pp. 345-352, 1980.

....supported by a gift from Intel Corporation and NASA grant NAGW419. In addition to routing speed, deadlock freedom is also a desired property of the data routing scheme. Store andforward and wormhole routing are two popular data routing schemes, and each has a different way to prevent deadlock [3][4]. Wormhole routing has the advantage that only the header flit needs to be buffered at the nodes for routing. Therefore wormhole is less sensitive to the long distance if the packet size is large [10] However, in many applications, the packet size is fixed and not very large. For example, in ....

....different segment lengths. Fig. 4 shows examples of good and bad alignment. 5 Deadlock Breaking Deadlock avoidance is an important issue in evaluating performance of data routing. There are many deadlock free routing algorithms, most of them require more hardware (buffers) and longer latency [3][4][8] Here we propose another algorithm called hill climbing (or side tracking, misrouting) to break deadlock situations. Hill climbing is similar to the Chaos Router[12] both are non minimal adpative routing. 5.1 Hill climbing (HC) algorithm In normal data routing, we always send packets in the ....

P. M. Merlin and P. J. Schweitzer, "Deadlock avoidance in store-and-forward networks I: Store-and-forward deadlock," IEEE Trans. on Commun., Vol. COM-28, no. 3, pp. 345-354, March 1980.

....deadlock freedom are essentially the same, because virtual cut through effectively reduces to store and forward routing when packets get blocked and buffered. Deadlock prevention for these cases is usually accomplished using structured buffer pools to create an acyclic buffer dependency graph [22, 43, 84]. However, in wormhole routed networks, packets are not buffered as whole units during routing. In this case, the cyclic dependency is for channels (vs. packet buffers) and requires the use virtual channels to break the deadlock [29] A good description of these deadlock prevention techniques can ....

P. Merlin and P. Schweitzer, Deadlock Avoidance in Store--and--Forward Networks --- I: Store--and--Forward Deadlock, IEEE Trans. on Communications, COM-28(3):345--354, March 1980.

.... which have to be maintained at each vertex or arc to allow deadlock free routing, and several techniques have been developed to design routing functions with a small number of buffers that avoid deadlocks by ordering buffers and allowing each message to use them in a monotonic increasing fashion ([17, 15, 19, 9, 1, 12, 11, 18, 3, 4, 8, 13] among the others) This idea results in the generation of a directed acyclic resource dependencies graph (DAG) thus preventing deadlock configurations. DAG based methods can be used with slight modifications both for packet and wormhole routing. In this paper a similar idea is considered, in ....

P.M. Merlin and P.J. Schweitzer. Deadlock avoidance in store-and-forward networks: Store and forward deadlock. IEEE Trans. on Commun., COM-28:345--352, March 1980.

....forms of communication by passing point topoint messages. This makes the cost of barrier synchronization and broadcasting very expensive. 2) Low latency communication. We used a new automatic routing scheme for the Tnet, which combines wormhole routing and a structured buffer pool algorithm[6]. This scheme achieves low latency, high throughput, and deadlock free communication between cells. By using the B net and S net, the latency of broadcast and barrier synchronization are also decreased. Each cell has a message controller (MSC) for fast message handling. As the grain size of ....

Philip M. Merlin, Paul J.Schweitzer "Deadlock avoidance in store-and-forward networks-I: store-andforward deadlock", IEEE Trans. Comm., COM-28, (3), pp.345-354, 1980. 4

....extreme example of unfairness, in which packets are stuck as they wait indefinitely in a queue while other packets (including new arrivals) are continually served. Many different types of deadlocks have been identified and studied, along with methods of preventing them (see, e.g. 1] 2] [3], 4] 5] 6] Figure 1 shows a simple example in which the cycle of nodes X, Y, and Z are in a deadlocked state. Note that X, Y, and Z can be any three adjacent nodes that form a cycle in a larger network of nodes. Proper congestion control is especially an important topic in emerging ....

P. M. Merlin and P. J. Schweitzer, "Deadlock Avoidance in Store-andForward Networks - I: Store-and-Forward Deadlock," IEEE Trans. Commun. , vol. COM-28, pp. 345-354, Mar. 1980.

....polynomial: O(n 2 mD) linear : O(nD) AG91] polynomial: O(nm log n mD) linear: O(n D) KOR95] polynomial: O(n 2 mD) logarithmic: O(logn D) Slide is a simple and efficient method for delivering tokens across an unreliable network. Like the Merlin Schweizer deadlock avoidance algorithm [MS80] it uses store and forward buffer hierarchies to control packet flow. However, the similarity ends here: slide allows packets the freedom to move in the network obliviously and permits deadlocks caused by individual packets that are delayed in the network for an indefinite periods of time. It ....

P. M. Merlin and P. J. Schweitzer. Deadlock avoidance in store-and-forward networks 1: Store-and-forward deadlock. IEEE Transaction on Communications, 28(3):345--354, March 1980.

....routing function is properly designed in order to avoid the occurrence of deadlocks. Several techniques have been developed to design deadlock free routing functions in which deadlocks are avoided by ordering the buffers and allowing each message to use them in a monotonically increasing fashion ([17, 15, 19, 9, 1, 12, 11, 18, 3, 4, 8, 13] among the others) As a consequence of the monotone usage of the buffers, resource dependencies are modeled by a directed acyclic graph (DAG) and this insures deadlock prevention. DAG based methods can be used with slight modifications both for packet and wormhole routing. In this paper a similar ....

P.M. Merlin and P.J. Schweitzer. Deadlock avoidance in store-and-forward networks: Store and forward deadlock. IEEE Trans. on Commun., COM-28:345--352, March 1980.

....because the queues of the message system are full. Obviously, deadlocks arise because the number of resources is finite. Many deadlock free routing algorithms have been developed for store andforward computer networks. Most of them require the use of central queues, restricting buffer allocation [7, 16, 19, 21, 28, 33]. These algorithms are also applicable to virtual cut through networks with central queues. Although algorithms that use central queues require less storage than those using edge buffers, central queues can become a bottleneck. So, algorithms that use edge buffers usually achieve a higher ....

P.M. Merlin and P.J. Schweitzer, "Deadlock avoidance in store-and-forward networks -- I: Store-and-forward deadlock," IEEE Trans. Commun., vol. COM28, no. 3, pp. 345--354, March 1980.

.... generating a protocol A 00 from an oblivious routing algorithm A that requires constant storage per node and uses routes of length O(log n) such as the routing algorithm for a binary tree network in which each node has two buffers, one for moving up the tree and one for moving down the tree [9]. However, this approach has the disadvantage of creating a bottleneck near the root of the tree. ....

P. M. Merlin and P. J. Schweitzer. Deadlock avoidance in store-and-forward networks-I: Store-and-forward deadlock. IEEE Transactions on Communications, 28(3):345--354, 1980.

....capacity in the nodes and the buffer management required to achieve these properties depend on whether wormhole [10] virtual cut through [22] or store and forward [44] routing is used. The requirements for store and forward and virtual cut through routing have been studied extensively [6, 7, 16, 29]. For concurrent use of all channels of a node, wormhole routing requires a buffer for each virtual channel, which is a more stringent requirement than what is required for virtual cut through and store and forward routing. For these routing techniques deadlock and livelock can be guaranteed ....

P. M. Merlin and P. J. Schweitzer. Deadlock Avoidance in Store--and--Forward Networks --- I: Store--and-- Forward Deadlock. IEEE Trans. on Communications, COM-28(3):345--354, March 1980.

....with an input and output buffer. Virtual channels may share physical communications links, but the buffers are 3 private. Virtual channels can be used to make routing deadlock free [5] The technique used is a generalization of deadlock prevention methods for packet switched networks [8, 22]. 4 ROMM Routing ROMM algorithms route each message from source to destination in k phases, k n. During each phase, a subset of the dimensions in the message s routing set are traversed. The dimensions traversed in each phase are determined by randomly selecting, for each message, k Gamma 1 ....

P. M. Merlin and P. J. Schweitzer. Deadlock Avoidance in Store--and--Forward Networks --- I: Store--and--Forward Deadlock. IEEE Trans. on Communications, COM-28(3):345--354, March 1980.

....not seem desirable to use a scheme in which minor changes in topology can degrade performance in such a major way. It is natural to ask whether there are more effective deadlock avoidance schemes. Many deadlock free routing algorithms have been developed, primarily for store and forward networks [MS80]. These algorithms are based on structured buffer pools. Buffers are partitioned into classes, and the assignment of buffers to messages is restricted in a way that prevents cycles. Unfortunately, structured buffer pool approaches require enough buffer capacity at each link to store multiple ....

P.J. Merlin and P.J. Schweitzer. Deadlock Avoidance in Store-andForward Networks -- I: Store-and-Forward Deadlock. IEEE Transactions on Communications, Vol. C-28, No. 3, March 1980.

....moves towards the destination. In general, the paths obtained subject to such Up Down routing are not the shortest paths that are possible in the network. Another approach to avoiding deadlocks consists of defining a partial order on the buffers (hierarchy of packet buffer pools) at each node [8, 9, 10, 21, 28, 29]. Most of these schemes avoid deadlocks by defining a partial order on the buffers and allocating them on various criteria like number of hops traveled, the route which the packet is traveling etc. Further, priorities in the order of service are defined among the buffers. For packet switching it ....

P. M. Merlin and P. J. Schweitzer. Deadlock Avoidance in Store-and-Forward Networks - I: Store-and-Forward Deadlock. IEEE Transactions on Communications, COM-28(3):345--354, March 1981.

....P1 s. Hence, it may be necessary for P1 to traverse the network again to decombine with all its packets. 2.2 Deadlock handling in the SSE A basic problem with recirculating networks with finite sized buffers is the possibility of deadlock. Classical deadlock avoidance schemes (see, for example, [MeSc80]) usually involve a fair amount of buffer space or a large number of virtual network channels. They are too involved to be implemented in hardware or firmware. In addition, buffer management schemes are usually designed to avoid packet collision. This is contrary to the principle of combining, ....

P. Merlin and P. Schweitzer, "Deadlock avoidance in store-and-forward networks --- I: store-and-forward deadlock," IEEE Trans. on Communications, Vol. COM28, No.3, 1980.

No context found.

P.M. Merlin and P.J. Schweitzer. Deadlock avoidance in store-and-forward networks: Store and forward deadlock. IEEE Trans. on Commun., COM-28:345--352, March 1980.

No context found.

P. M. Merlin and P. J. Schweitzer. Deadlock avoidance in store-and-forward networks 1: Store-and-forward deadlock. IEEE Transaction on Communications, 28(3):345--354, March 1980.

No context found.

P.M. Merlin and P. J. Schweitzer, "Deadlock avoidance in store-andforward networks--I: Store and forward deadlock," IEEE Trans. Commun., vol. COM-28, pp. 345-352, Mar. 1980.

No context found.

P. Merlin and P. Schweitzer. Deadlock avoidance in store-and-forward networks-I : store-and-forward deadlock. IEEE Transactions on Communication, 28:325, 1980.

No context found.

P. Merlin and P. Schweitzer. Deadlock avoidance in storeand -forward networks -- I: Store-and-forward deadlock. IEEE Transactions on Communications, COM-28:345--354, March 1980.

First 50 documents

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