Bounded Site Failures: An Approach to Unreliable Grid Environments.
Joaquim Gabarró, Alina García, Maurice Clint, Peter Kilpatrick, A L Stewart · Research Portal (Queen's University Belfast) · 2007
The abstract behaviour of a grid application management system can be modelled as an Orc expression in which sites are called to perform sub-computations. An Orc expression specifies how a set of site calls are to be orchestrated so as to realise some overall desired computation. In this paper evaluations of Orc expressions in untrusted environments are analysed by means of game theory. The set of sites participating in an orchestration is partitioned into two distinct groups. Sites belonging to the first group are called angels: these may fail but when they do they try to minimize damage to the application. Sites belonging to the other group are called daemons: when a daemon fails it tries to maximise damage to the application. Neither angels nor daemons can fail excessively because the number of failures, in both cases, is bounded. When angels and daemons act simultaneously a competitive situation arises that can be represented by a so-called angel—daemon game. This game is used to model realistic situations lying between over-optimism and over-pessimism.