Garbage collector assíncrono em simulação distribuída

Abstract

Distributedsimulationsneedsynchronizationstrategiestoavoidlocalcausalityconstraint(LCC) violations, which happen when the event execution order disrespects their chronological order. Conservativestrategiesavoidsuchviolationsaltogether. Optimisticstrategiesallowcomponents to execute independently, which makes them susceptible to LCC violations. Proper synchro- nization is ensured by saving previous consistent states in checkpoints so that the simulation can rollback to a consistent state whenever a violation occurs. The process of saving checkpoints can follow various criteria, but they will always increase in number as the simulation progresses. This causes the need to remove checkpoints as the simulation evolves, to prevent memory usage to grow unboundedly. Optimistic synchronization splits into coordinated, in which control messages are used; unco- ordinated, which doesn’t have control messages whatsoever; and communication induced, in whichtheinformationusedbythecoordinationprotocolispiggybackedinapplicationmessages. In coordinated synchronization, checkpoint removal is a trivial process. This is not the case for uncoordinated synchronization and communication induced synchronization, which require the aid of algorithms whose sole purpose is checkpoint removal; they are called garbage collectors, and can also be coordinated, uncoordinated or communication induced. Asynchronous garbage collectors (the ones that are uncoordinated or communication induced) havetheadvantageofnothavingcoordinationoverhead,whichpreservessomeoftheadvantages of not using a coordinated synchronization in the first place. This work proposes a heuristic garbage collection strategy. The proposed strategy uses metrics in order to compute which checkpointsmustberemovedatanygiventime. ThesolutionwasimplementedintheDistributed Co-Simulation Backbone, a distributed heterogeneous simulation architecture that still doesn’t have a garbage collection implementation.