Enhancing Performance and Energy Efficiency for Hybrid Workloads in Virtualized Cloud Environment
AbstractVirtualization has attained mainstream status in enterprise IT industry. Despite its widespread adoption, it is known that virtualization also introduces non-trivial overhead when executing tasks on a virtual machine (VM). In particular, a combined effect from device virtualization overhead and CPU scheduling latency can cause performance degradation when computation intensive tasks and I/O intensive tasks are co-located on a VM. Such an interference causes extra energy consumption, as well. In this paper, we present Hylics, a novel solution that enables efficient data traverse paths for both I/O and computation intensive workloads. This is achieved with the provision of in-memory file system and network service at the hypervisor level. Several important design issues are pinpointed and addressed during our prototype implementation, including efficient intermediate data sharing, network service offloading, and QoS-aware memory usage management. Based on our real-world deployment on KVM, Hylics can significantly improve computation and I/O performance for hybrid workloads. Moreover, this design also alleviates the existing virtualization overhead and naturally optimizes the overall energy efficiency.
EXISTING SYSTEM
With the major focus on resolving the I/O bottleneck, the existing work does not provide comprehensive evaluations on hybrid workload performance in cloud environments. Hybrid workloads may experience performance degradation in multiple aspects, including I/O and computation performance. The impact remains largely unexplored, and a solution is yet to be developed for common cloud services demanding both data processing and transmission.
PROBLEM DEFINITION
• Existing virtualization technologies, such as Xen1 and KVM2, also introduce non-trivial overhead when executing tasks on a virtual machine (VM).
• This leads to longer and unstable task completion time for computation-intensive applications .
• An overhead also causes self interference for hybrid workloads that involve both computation and I/O intensive tasks.
• Different from cross- VM interference3, self interference happens within a VM when the I/O handling process of the VM is interfered or even starved by other processes inside the VM.
PROPOSED SYSTEM
In this paper, We for the first time performed a comprehensive measurement study to quantify the impact of self interference with hybrid workloads. Hybrid workloads such as transcoding and streaming tasks experience up to 32.1% reduction of network throughput, up to 32.5% reduction of computation performance. Motivated by the measurement results and an in-depth analysis, we present Hylics, a novel virtualization architecture that jointly optimizes I/O and computation performance for hybrid workloads. The insight of the Hylics design is to shorten the data traverse path for both processing and transmission. Meanwhile, it also decouples I/O and computation operations for cloud VMs. In particular, Hylics stores cloud applications’ data in the in-memory file system at the hypervisor level. By doing this, the data traverse path now originates, or ends, at hypervisor-level memory space. The design also shifts VM’s network operations to the hypervisor layer. The possible self interference between I/O and computation is therefore minimized, enabling near bare-metal networking performance and enhanced computation performance.
CONCLUSION
Thus this paper closely examined the self interference from real-world applications in virtualized environments. To jointly optimize performance and energy efficiency for hybrid workloads in cloud environments, we designed and developed Hylics, a novel protocol-independent solution that leverages the hypervisor-level in-memory file system sharing. We implemented a prototype of Hylics in KVM and evaluated the overall performance through real-world workloads, which indicates that such a design can largely improve I/O performance and accelerate computation tasks in the presence of the self interference. The energy efficiency of the underlying server is also enhanced. In the future work, we plan to implement Hylics-based solutions on other virtualization technologies. Since Hylics significantly minimizes the self interference, we will also revisit the VM resource allocation issues to help cloud providers to achieve better service performance and cost efficiency.