LIP6 1999/006
- Thesis
De l'amélioration des performances par la réplication des objets :
une approche par estampillage - J.-Ph. Lesot
- 174 pages - 03/11/1999- document en - http://www.lip6.fr/lip6/reports/1999/lip6.1999.006.ps.gz - 413 Ko
- Contact : lesot (at) nullsrc.lip6.fr
- Ancien Thème : SRC
- Keywords : distributed systems, objects'systems, performance, coherency, active replication, synchronization
- Publisher : Denis.Poitrenaud (at) nulllip6.fr
Speeding up distributed application is a significant quality criteria for users. Communication channels have associated propagation time as well as latency which slow clients down. The use of wireless appliances connected on a discontinuous basis or with a low and/or irregular bandwidth add even more constraints to the system. Our main goal is to improve server performance in an distributed objects' system. Our main contributions are (i) a framework, a partial architecture design of an objects' system using active replication, and (ii) an invocation scheduling stamp protocol. Firstly, the invocation scheduling stamp protocol allows a query to carry all the necessary information required to respect a delivery order fixed by a coherency model. The implementation of this model is simply done using our protocol to establish an order between queries by identifying and
stamping them. This method hides the query propagation, spooling, reordering and delivery problems to application. Thus, we hope to increase the use of a strictly necessary coherency model, or even the use of application specific model, by suggesting a simple and efficient way. Secondly, in our framework, each manager provides a focused functionality, independently of other managers, allowing each one to take advantage of its environment and to be easily interchangeable. Asynchronous communications between those different managers are favoured as well as local interactions. Finally, we have built a prototype of the framework. Its shows the low cost of the invocation scheduling stamp protocol and how easy it is to implement coherency models and synchronization mechanisms with it.
stamping them. This method hides the query propagation, spooling, reordering and delivery problems to application. Thus, we hope to increase the use of a strictly necessary coherency model, or even the use of application specific model, by suggesting a simple and efficient way. Secondly, in our framework, each manager provides a focused functionality, independently of other managers, allowing each one to take advantage of its environment and to be easily interchangeable. Asynchronous communications between those different managers are favoured as well as local interactions. Finally, we have built a prototype of the framework. Its shows the low cost of the invocation scheduling stamp protocol and how easy it is to implement coherency models and synchronization mechanisms with it.