A case for dynamic memory partitioning in data centers
Proceedings of the Second Workshop on Data Analytics in the Cloud
Adaptive Online Compression in Clouds--Making Informed Decisions in Virtual Machine Environments
Journal of Grid Computing
| Hi-index | 0.00 |
IaaS clouds have become a promising platform for scalable distributed systems in recent years. However, while the virtualization techniques of such clouds are key to the cloud's elasticity, they also result in a reduced and less predictable I/O performance compared to traditional HPC setups. Besides the regular performance degradation of virtualized I/O itself, it is also the potential loss of I/O bandwidth through co-located virtual machines that imposes considerable obstacles for porting data-intensive applications to that platform. In this paper we examine adaptive compression schemes as a means to mitigate the negative effects of shared I/O in IaaS clouds. We discuss the decision models of existing schemes and analyze their applicability in virtualized environments. Based on an evaluation using XEN, KVM, and Amazon EC2, we found that most decision metrics (like CPU utilization and I/O bandwidth) are displayed inaccurately inside virtual machines and can lead to unreasonable levels of compression. As a remedy, we present a new adaptive compression scheme for virtualized environments which solely considers the application data rate. Without requiring any calibration or training phase our adaptive compression scheme can improve the I/O throughput of virtual machines significantly as shown through experimental evaluation.