T. Johnson and K. Jeong. Hierarchical matrix timestamps for scalable update propagation. submitted to the 10th Workshop on Distributed Algorithms, June 1996.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....which require causal broadcast and therefore do not scale well. These solutions include [12] in which nodes are grouped in hierarchically structured clusters, and [13] in which nodes are organized in a Daisy architecture. A solution based on Hierarchical Matrix Timestamps (HMT) is proposed in [14]. The HMT stores information about other nodes in the same domain and summarizes information about other domains. But this technique is specially adapted to update propagation in replicated databases using a weak consistency algorithm 3 and is not suitable for causal communication. An original ....

....[21] Causality is only maintained on each individual bus, also called a domain of causality. The idea of splitting the set of servers is not new, but the published solutions either use vector clocks and broadcast [12, 13] or use matrix clocks but only reduce the message timestamp size [14, 15]. A quite similar solution based on group composition was used in [22] but the authors only prove that if the groups topology is a tree the end to end ordering is respected, but not the opposite. Moreover the ordering algorithm was not given, and no implementation exists. In our paper we prove ....

T. Johnson and K. Jeong. Hierarchical matrix timestamps for scalable update propagation. submitted to the 10th Workshop on Distributed Algorithms, June 1996.

....is especially serious in singlemaster systems in which the master replica propagates updates to all the slaves. This problem, fortunately, is alleviated easily. One solution is to organize the replicas in a tree form, place the master at root, and let updates ripple down the tree from the root [15, 80, 1, 34]. It cuts the propagation delay from O(N) to O(logN) N is the number of replicas) and reduces the load on the master (and all other replicas) to a constant level. Multi master systems such as Roam [65] Usenet [69] and Active Directory [49] can push this idea further and connect replicas in a ....

T. Johnson and K. Jeong. Hierarchical matrix timestamps for scalable update propagation. http://w4.lns.cornell.edu/jeong/index-directory/hmtpaper. ps, June 1996.

....which require causal broadcast and therefore do not scale well. These solutions include [14] in which nodes are grouped in hierarchically structured clusters, and [19] in which nodes are organized in a Daisy architecture. A solution based on Hierarchical Matrix Timestamps (HMT) is proposed in [22]. The HMT stores information about other nodes in the same domain and summarizes information about other domains. But this technique is specially adapted to update propagation in replicated databases using a weak consistency algorithm 2 and is not suitable for causal communication. An original ....

....[1] Causality is only maintained on each individual bus, also called a domain of causality. The idea of splitting the set of servers is not new, but the published solutions either use vector clocks and broadcast [19] 14] or use matrix clocks but only reduce the message timestamp size [22][11] Our solution reduces both the message timestamp size and the control information size on servers. In addition, the modularity of the architecture allows it to be adapted to any physical or logical topology. A domain of causality is a group of servers in which the causal order is respected. ....

T. Johnson and K. Jeong. Hierarchical Matrix Timestamps for Scalable Update Propagation, Unpublished work, submitted to the 10 th Workshop on Distributed Algorithms, June 1996.

....systems in which the master replica is responsible for propagating updates to all the other replicas. This problem, fortunately, is alleviated relatively easily. One solution is to connect replicas in a tree structure, place the master at root, and let updates ripple down the tree from the root [14, 76, 1, 32]. It cuts the propagation delay from O(N) to O(log N) N is the number of replicas) and reduces the load on the master to a constant level. Multi master systems such as Roam [62] Usenet [66] and Active Directory [47] can push this idea further and connect replicas in a tree structure supplemented ....

T. Johnson and K. Jeong. Hierarchical matrix timestamps for scalable update propagation. http://w4.lns.cornell.edu/~jeong/index-directory/hmt-paper.ps, June 1996.

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