A Single System Image Supporting Distributed Objects

Abstract

Nowadays, parallel systems, such as networks of heterogeneous workstations, are widely used to solve distributed applications. An easy way to implement suCh applications is the use of a distributed memory which is shared among all machines participating ii) the application. In such environment, failures, such as machine crashes, are harmful because of the possibly lost data due to the failure. In most cases the application is to be rest&rted from scratch. If the application is a computational intensive one and if the handled data must remain alive to survive the application, failures must be handled dynamically. Our solution to such problem adds an additional layer, called the high availability l(lyer, to the shared memory architecture. Such layer includes mechanisms to handle the failures dynamically. To implement such layer, the protocols introduced by the TR.ansis1 system are used as well as the shared memory supported by the LIPS2 system. Hence, we named the resulting system TRlPS. TRIPS is a way to achieve a unified view'! to the shared memory among all nodes in the parallel machine in spite of the failures. Its aim is to establish a pro ocol to achieve a Single System Image.