Resource-aware Collaborative Allocation for CPU-FPGA Cloud Environments

Abstract : Cloud Warehouses have been exploiting CPU-FPGA environments to accelerate multi-tenant applications to achieve scalability and maximize resource utilization. In this scenario, kernels are sent to CPU and FPGA concurrently, considering available resources and workload characteristics, which are highly variant. Therefore, we propose a multi-objective optimization strategy to improve resource provisioning in CPU-FPGA environments. It is based on the Genetic Multidimensional Knapsack solution and can be tuned to minimize makespan or energy. Our strategy provides similar results as the optimal Exhaustive Search, but with feasible execution time, while presenting 77% energy savings with 39% lower makespan than the commonly-used First-Fit strategy.

 EXISTING SYSTEM :

 ? In, various case studies have been presented comparing GPU vs. FPGA implementations. ? There seems to be no “one fits all” solution, but the benefits and drawbacks are manifold: GPUs benefit from higher clock frequencies than FPGAs, due to their custom layout. ? On the other hand, programming GPUs efficiently can be as tedious as hardware development in VHDL and their massive power consumption prevent them from being used on mobile robots. ? FPGAs offer a higher degree of freedom in optimizing memory architectures, data types and pipeline structures at lower power consumption.

 DISADVANTAGE :

 ? Domain-specific High-Level Synthesis (HLS) provides programming abstractions to ease the problem specification and thus productivity. ? A solution to the modulo scheduling problem can be found using only the non-parametric constraints. ? The major problem still facing medical images is the poor Signal-to-Noise Ratio (SNR) due to limited dosage and exposure for health reasons. ? Solving above-mentioned scientific problems requires the expertise of different specialists, including domain experts (i.e., natural scientists), mathematicians, software as well as hardware engineers.

 PROPOSED SYSTEM :

 • In order to evaluate the impact of pruning on the accuracy of detected corners, we use the metrics named precision and recall as proposed in. • The value of recall measures the number of correct matches out of the total number of possible matches, and the value of precision measures the number of correct matches out of all matches returned by the algorithm. • The computation of the nearest neighbor for the purpose of feature matching is the most time-consuming part of the complete recognition and localization algorithm. • This algorithm performs a heuristic search and only visits a fixed number of leaves resulting in an actual nearest neighbor, or a data point close to it.

 ADVANTAGE :

 ? It captures domain knowledge in a compact and intuitive language and employs source-to-source translation combined with various optimizations to achieve an excellent productivity paired with performance portability. ? To reduce the foreseen performance/productivity gap of upcoming exascale platforms, a unique, tool-assisted co-design approach specific for the domain of multigrid methods based on stencil computations is developed within ExaStencils. ? Whereas, GPUs are high-performance devices, which are well suited for 3D imaging when floating-point calculations are needed. ? The proposed techniques lead to significant advantages with respect to productivity, portability, and performance.