Energy-Aware Real-time Tasks Processing for FPGA Based Heterogeneous Cloud
ABSTARCT :
Cloud computing is becoming an popular model of computing. Due to the increasing complexity of the cloud service requests, it often exploits heterogeneous architecture. Moreover, some service requests (SRs)/tasks exhibit real-time features, which are required to be handled within a specified duration. Along with the stipulated temporal management, the strategy should also be energy efficient, as energy consumption in cloud computing is challenging. In this paper, we have proposed a strategy, called ``Efficient Resource Allocation of Service Request" (ERASER) for energy efficient allocation and scheduling of periodic real-time SRs on cloud platform. Our target cloud platform is consist of Field Programmable Gate Arrays (FPGAs) as Processing Elements (PEs) along with the General Purpose Processors (GPP). We have further proposed, a SR migration technique to service maximum SRs. Simulation based experimental results demonstrate that the proposed methodology is capable to achieve upto 90% resource utilization with only 26% SR rejection rate over different experimental scenarios. Comparison results with other state-of-the-art techniques reveal that the proposed strategy outperforms the existing technique with 17% reduction in SR rejection rate and 21% less energy consumption. Further, the simulation outcomes have been validated on a real test-bed based on Xilinx Zynq SoC with benchmark tasks.
EXISTING SYSTEM :
? Most existing works on CPU DVFS only consider scaling the CPU voltage or frequency alone, while existing works have shown that the GPU core voltage, GPU core frequency, and GPU memory frequency are the major factors that affect the dynamic GPU power.
? To maximize the energy efficiency brought by GPU DVFS, some existing models generally adopted machine learning methods to clarify the application patterns and predict the effects of DVFS on performance and energy.
? The energy consumption of the additional CPU cores, if any, can be modeled as static energy if they are idle, or be handled by existing CPU energy management techniques if they are used to run CPU jobs.
DISADVANTAGE :
? Acceleration in execution of a service request on such heterogeneous architecture is a challenging issue in cloud.
? To tackle the energy minimization problem, we need to accurately understand the GPU performance model and power model.
? We then extend the offline EDL algorithm to the online problem, where we combine GPU DVFS and dynamic resource sleep (DRS).
? One challenge in this scheduling problem is to achieve a good balance between dynamic energy consumption (which prefers low voltage/frequency and long execution time) and static energy consumption (which prefers high voltage/frequency and low execution time).
PROPOSED SYSTEM :
• They proposed a variant of the HEFT algorithm and discussed the cases with and without using DVFS.
• Based on the basic hardware information, a series of studies proposed several analytical models to estimate different degrees of memory traffic and computational workload of a kernel, and then calculate the execution cycles according to different cases.
• We propose the GPU-specific DVFS power and performance models, and derive the appropriate GPU voltage/frequency setting through the mathematical optimization.
• As for task scheduling and mapping on hybrid servers, we first propose the offline solution EDL to it.
ADVANTAGE :
? A user can choose particular hardware, storage and processing platforms in a cloud, based on the performance requirements through a service request (SR).
? Due to the wide variety of performance demands from the users, cloud service providers often need to be very flexible in terms of service delivery.
? The performance and efficiency of the homogeneous CPU based cloud is insufficient to match the requirements of modern cloud servers.
? FPGAs provide the support of accelerator for modern high performance cloud computing platforms.
? It can be observed that TPA-STA performance is poor than ERASER in terms of rejection rate as well as the energy consumption, as the L increases.
|