A Framework for Performance Evaluation of Decentralized Eventual Consistency Algorithms

Mehdi Ahmed-Nacer,Pascal Urso

doi:10.4108/eai.30-6-2017.152755

Abstract

Eventual Consistency (EC) model is adopted by numerous large-scale distributed systems. To ensure performance and scalability, this model allows any replica to accept updates without remote synchronization. Nowadays, many EC algorithms are developed to control the behavior of the replicated data in the face of concurrent updates. Among them, those using a central server to order the updates, while others support the decentralization. In this paper, we focus on decentralized EC algorithms. Suitability of such algorithms under users and devices constraints such as execution time, memory requirements, messages size and quality of the result remains to be investigated under different conditions. Evaluate such algorithms in different context and under different parameters require a framework. In this paper, we propose a generic framework designed to evaluate diferent decentralized EC algorithms, in diferent context by controlling diferent parameters. Our framework provides a generic simulator that generates a runnable data following diferent parameters.

Highlights

Replication is a key feature in any large distributed systems
We present the experimental results retrieved through our framework, by using synchronous corpus
Even if the framework supports Convergent Replicated Data Type (CvRDT) and Commutative Replicated Data Type (CmRDT) (Section ), we present as an example in this paper only results of CmRDT

Summary

Introduction

Replication is a key feature in any large distributed systems. When the replicated data are mutable, the consistency between the replicas must be ensured. A different model of consistency can be established. In the strong consistency model (aka atomic or linear consistency), a mutation seems to occur instantaneously on all replicas. Within a strong consistency model, integrity constraint on the data manipulated can be ensured through transactions. The CAP theorem [6, 10] states that it is impossible to achieve simultaneously strong consistency (C), availability (A) and to tolerate network partition (P)

Results

Discussion

Conclusion