Fault-Tolerant Replication Based on Fragmented Objects

Abstract

Abstract. This paper describes a novel approach to fault-tolerance in distributed object-based systems. It uses the fragmented-object model to integrate replication mechanisms into distributed applications. This approach enables the use of customised code on a per-object basis to access replica groups and to manage consistency. The addition of fault tolerance to the infrastructure has only little overhead, is fully transparent for clients, and does not require internal modifications to the existing middleware. Semantic annotations at the interface level allow the developer to customise the provision of fault tolerance. Operations can be marked as read-only to allow an execution with weaker ordering semantics or as parallelisable to allow true multithreaded execution. A code-generation tool is provided to automatically produce object-specific fragment code for client access and for replica consistency management, taking into account the annotations, the interface specification, and the non-replicated implementation. A further contribution of our code-generation approach is the support of deterministic multithreading in replicated objects.

The Architecture of the XtreemOS Grid Checkpointing Service

“…Furthermore, it distributes the load, induced by XtreemGCP over the whole grid. We use virtual nodes [21] to make job checkpointers and job managers highly available via service replication.…”

Using Object Replication for Building a Dependable Version Control System

“…The FTflex object replication infrastructure [20] uses the fragmented-object model [14,11,19] provided by the Aspectix middleware to integrate replication mechanisms into distributed applications. From an abstract point of view, a fragmented object is defined by its identity, interface, functionality, and state.…”

“…This code substitutes the static stub and skeleton that is commonly used in distributed object middleware. The FTflex replication infrastructure [20] combines the fragmented object model with semantic annotations in order to customize and optimize replication mechanisms. In this paper, we use a specific complex application-a distributed version control system-to assess the replication mechanisms provided by that replication infrastructure.…”

Revisiting Deterministic Multithreading Strategies

“…Multiple deterministic scheduling algorithms [1,2,3,11] have been proposed in the past. We included all of them in FTflex [12] the replication framework for our Java-based and CORBA-compatible Aspectix middleware. In the first part of this paper we give a survey on existing algorithms.…”