The shared data space model has proven to be an effective paradigm for building distributed applications. However, building an efficient distributed implementation remains a challenge. A plethora of different implementations exists. Each of them has a specific policy for distributing data across nodes. Often, these policies are tailored to a specific application domain. Thus, those systems may often perform poorly with applications extraneous to their domain. In this paper, we propose that implementations of a distributed shared data space system should provide mechanisms for tailoring data distribution policies. Through this flexibility the shared data space system can cope with a wide spectrum of application classes. The need for this flexibility is illustrated by experiments which show that there is no single distribution policy that works well in all cases.
|Titel||Euro-Par 2004 Parallel Processing (Proceedings 10th International Conference, Pisa, Italy, August 31-September 3, 2004)|
|Redacteuren||M. Danelutto, M. Vanneschi|
|Plaats van productie||Berlin|
|ISBN van geprinte versie||3-540-22924-8|
|Status||Gepubliceerd - 2004|
|Naam||Lecture Notes in Computer Science|
|ISSN van geprinte versie||0302-9743|