DONUT: efficient data localization
Having a number of copies adapted to the popularity of the data is not enough. It is also necessary to be able to locate the data efficiently. Distributed systems large scale bring together thousands of nodes distributed worldwide. The data is distributed over a logical network according to their identifier. To provide effective localization, systems must maintain shortcuts in their graph logic. However, to create efficient shortcuts, the nodes must have information about the topology of the overlay. In the case of heterogeneous distribution of nodes, obtaining such information is not easy. Moreover, due to the presence of churn, the topology can evolve quickly, making the collected information obsolete. To deal with this problem DONUT constructs a map local which gives an approximation of the distribution of nodes, thus allowing an estimation of the distance between the nodes in the graph. The ratings we have performed with real latencies and churn traces have shown that our map can increase routing efficiency by more than 20% compared to state-of-the-art techniques. art.
Operation of DONUT:
DONUT is a mechanism that provides on each node, a global map that makes it possible to estimate the distance with the other nodes in the graph. The main idea is to build, on each node, a fuzzy view of the distribution of nodes on the whole system. This card is then used to build effective long links. With DONUT, creating long links is inexpensive, as it is done locally using the map. In case of modification due to churn, the local map gradually adapts to reflect the new density distributions, allowing the nodes to replace long links that have become obsolete with new ones. Unfortunately, there is no algorithm to build long links based on the use of a map that approximates the global distribution of nodes in the system The contributions modified by DONUT are: a distributed algorithm providing each node with a map global distribution of nodes in the system. an algorithm that builds long links. Finally, in order to allow efficient localization of data while preserving their semantic relations, we have designed DONUT. DONUT builds on each node a global fuzzy map of density distribution which allows to set up efficient routing. Also, this map may be useful for other important mechanisms for distributed systems, such as load balancing, estimating the size of the system, monitoring the system, or the special routing mechanisms. Performance is an important criterion of data replication mechanisms. But it is also necessary to take into account the consistency of the data. Often, a consistency/performance trade-off is necessary.
Data consistency management:
In computer science, consistency is the ability for a system to reflect on the copy of data the modifications occur on other copies of this data. This concept is mainly used in three computer fields:
file systems, databases, and memories shared. A constraining consistency model allows an intuitive behavior and simplifies the understanding of the behavior of programs, but consistency models weak or relaxed often improve the performance, it is up to the programs to ensure the consistency of the data examined is necessary. replication improves fault tolerance and the performance of storage systems. But the existence of several copies of the same datum poses problems of consistency during updates. For example, transactions parallels on replicated databases have overhead importance, due to the need to control competition and the occurrence of conflicts that could lead to the abandonment of transactions
You have observed very interesting details! ps decent web site.Money from blog