QoS in Cloud Storage

Cloud computing uses virtualization techniques to cater to multiple concurrent workloads belonging to different users on a single physical system. The additional layer of abstraction brought in by virtualization adds undesirable performance overheads for the applications’ performance. This issue is more relevant for latency sensitive disk I/O workloads where the co-hosted applications contend for limited I/O resources for their functionality. It is pertinent that the performance SLAs of such applications are satis ed through intelligent scheduling and allocation of disk resources. Additionally, keeping in mind that server consolidation in Cloud data-centers is an important goal of virtualization, we emphasize the need for optimal meta-scheduling techniques which can ensure increased utilization of disk resources for I/O applications while maintaining application performance at the same time. This work proposes novel scheduling frameworks for deciding¬† placement of applications on servers and managing disk resources to achieve these goals. PriDyn disk scheduler facilitates differentiated services for applications on a single host. This scheduler was extended to PCOS meta-scheduling framework to ensure proper workload combinations on the hosts for achieving consolidation along with desired QoS. In this work we have focused on providing performance guarantees for I/O intensive workloads executing in a virtualized Cloud data-center. We have proposed PriDyn, a latency-aware disk scheduler for the virtualized host that can perform dynamic allocation of disk resources to the various I/O applications. We performed extensive experiments to demonstrate the benefits of the proposed scheduler for real world Cloud data-centers. This idea was further extended to a data-center level meta-scheduler PCOS which aims to maximize the utilization of disk resources without adversely affecting the performance of the applications scheduled on the systems. PCOS intelligently selects optimal workloads combinations for the servers and achieves appreciable enhancements in disk I/O performance. Experimental validations performed on real-world I/O traces con rm that our framework can enable QoS guarantees along with efficient resource utilization for Cloud data-centers. Further, these algorithms were implemented in the CloudSIM software to support differentiated disk I/O services.

  • Nitisha Jain, J. Lakshmi, PriDyn : Enabling Differentiated I/O Services in Cloud using Dynamic Priorities”, IEEE Transactions on Services Computing (Special Issue on Cloud Computing), vol. PP, no. 99, 2014.(PDF)
  • Nitisha Jain, J. Lakshmi, PriDyn : Framework for Performance Specific QoS in Cloud Storage”, Proceedings of the 7th IEEE International Conference on Cloud Computing (IEEE CLOUD 2014), June 27 – July 2, 2014, Alaska, USA.(PDF)
  • N. Jain and J. Lakshmi, “PCOS: Prescient Cloud I/O Scheduler for Workload Consolidation and Performance,” 2015 International Conference on Cloud Computing and Big Data (CCBD), Shanghai, 2015, pp. 145-152. doi: 10.1109/CCBD.2015.24 (PDF)(Talk)
  • N. Jain, N. Grozev, J. Lakshmi and R. Buyya, “PriDynSim a Simulator for Dynamic Priority Based I/O Scheduling for Cloud Applications,” 2015 IEEE International Conference on Cloud Computing in Emerging Markets (CCEM), Bangalore, 2015, pp. 8-15. doi: 10.1109/CCEM.2015.17 (PDF)*
  • Performance Specific I/O Scheduling Framework for Cloud Storage, Nitisha Jain MSc(Engg) Thesis, 2015 (PDF)

* This work was completed under the aegis of MoU with CLOUDS lab of University of Melbourne.