Section: New Results
Cloud Storage Trade-Offs: Consistency and Self-Adaptiveness
Cost-aware consistency management in the cloud
Participants : Houssem-Eddine Chihoub, Shadi Ibrahim, Gabriel Antoniu.
With the emergence of cloud computing, many organizations have moved their data to the cloud in order to provide scalable, reliable and highly available services. To meet ever growing user needs, these services mainly rely on geographically-distributed data replication to guarantee good performance and high availability. However, with replication, consistency comes into question. Service providers in the cloud have the freedom to select the level of consistency according to the access patterns exhibited by the applications. Most optimizations efforts then concentrate on how to provide adequate trade-offs between consistency guarantees and performance. However, as the monetary cost completely relies on the service providers, in [20] we argue that monetary cost should be taken into consideration when evaluating or selecting a consistency level in the cloud. Accordingly, we define a new metric called consistency-cost efficiency. Based on this metric, we present a simple, yet efficient economical consistency model, called Bismar, that adaptively tunes the consistency level at run-time in order to reduce the monetary cost while simultaneously maintaining a low fraction of stale reads. Experimental evaluations with the Cassandra cloud storage on a Grid'5000 testbed show the validity of the metric and demonstrate the effectiveness of the proposed consistency model.
Analysis of the impact of consistency mangement on energy consumption
Participants : Houssem-Eddine Chihoub, Shadi Ibrahim, Gabriel Antoniu.
Energy consumption within datacenters has grown exponentially in recent years. In the era of Big Data, storage and data-intensive applications are one of the main causes of the high power usage. However, few studies have been dedicated to the analysis of the energy consumption of storage systems. Moreover, the impact of consistency management has never been investigated in spite of its high importance. In this work, we address this particular issue. We investigate the energy consumption of application workloads with different consistency models. Thereafter, we leverage the observations about power and the resource usage with every consistency level in order to provide insight into energy-saving practices. In this context, we introduce adaptive configurations of the storage cluster according to the used consistency level. Our experimental evaluations on Cassandra deployed on Grid'5000 demonstrate the existence of the inevitable tradeoff between consistency and energy saving. Moreover, they show how reconfiguring the storage cluster can lead to energy saving, enhanced performance, and better consistency.
Chameleon: customized consistency by means of application behavior modeling
Participants : Houssem-Eddine Chihoub, Gabriel Antoniu.
Multiple Big Data applications are being deployed worldwide to serve a very large number of clients nowadays. These applications vary in their performance and consistency requirements. Understanding such requirements at the storage system level is not possible. The high level semantics of an application is not exposed at the system level. In this context, the consequences of a stale read are not the same for all types of applications.
In [28] , we focus on managing consistency at the application level rather than at the system level. In order to achieve this goal, we propose an offline modeling approach of the application access behavior that considers its high–level consistency semantics. Furthermore, every application state is automatically associated with a consistency policy. At runtime, we introduce the Chameleon approach that leverages the application model to provide a customized consistency specific to that application. Experimental evaluations show the high accuracy of our modeling approach exceeding 96% of correct classification of the application states. Moreover, our experiments conducted on Grid'5000 show that Chameleon adapts, for every time period, according to the application behavior and requirements while providing best-effort performance.