Software Diversity-Based Active Replication as an Approach for Enhancing the Performance of Advanced Simulation Systems

Abstract

Active replication has been widely explored to achieve fault tolerance and to improve system availability, especially in service oriented applications. In this paper we explore software diversity-based active replication in the context of advanced simulation systems, with the aim at improving the timeliness for the production of simulation output. Our proposal is framed by the High-Level-Architecture (HLA), namely a middleware based standard for simulation package interoperability, and results in the design and implementation of an Active Replication Management Layer (ARML) targeted to off-the-shelf SMP computing systems. This layer can be interposed in between each simulator instance and the underlying HLA middleware component, in order to support the execution of diversity-based active replicas of a same simulation package in a totally transparent manner. Beyond presenting the replication framework and the design/implementation of ARML, we also report the results of an experimental evaluation on a case study, quantifying the benefits from our proposal in terms of both simulation execution speed and performance guarantees vs tunable software parameters. (Free software releases of ARML can be found at the URL ).

Automatic Algorithm Selection for Complex Simulation Problems

“…Quaglia [267,268] realized a simple form of continuous algorithm selection for speeding up single PDES runs: two replications are executed by different configurations in parallel, and the results of the fastest replicated LP are sent to both replications of the recipient. In other words, this approach employs a parallel portfolio approach in which the two algorithms of the portfolio interact with each other, i.e., they work together to solve the simulation problem.…”

“…Another approach from the domain of PDES, active replication, improves the speed of a single execution by letting several algorithms run in parallel per processor, each time using the result of the fastest one while dismissing those of the others [267,268]. Multiple replications of PDES runs may also be executed on the same resources in parallel, which achieves additional speedups [23].…”

A Three-Phases Byzantine Fault Tolerance Mechanism for HLA-Based Simulation

“…The concept of the decoupled federate architecture was also used implicitly in many other works. For instance, the FT components proposed in [14] and the ARML proposed in [15] are the middlewares which control the communication between a federate and its DRC. A service oriented HLA RTI (SOHR) [16] has been proposed recently.…”

“…The replicated objects introduced in [22] are treated as individual objects in the simulation execution, which are similar to the replicated federates in the traditional federate architecture. In contrast, the concept of the decoupled federate architecture was used implicitly in [14], [15]. The federate replication structure in [14] can be viewed as an example of the Barrier federate replication structure.…”

“…Therefore, comparing the RTI services invoked by replicas of the same federate, as reported in [14], is not enough for Byzantine failure detection. The federate replication structure in [15] can be viewed as an example of the No-Barrier federate replication structure. However, it focuses on performance enhancement issues rather than fault tolerance.…”