Anytime Algorithms for GPU Architectures

Mangharam, Rahul; Saba, Aminreza Abrahimi

doi:10.1109/rtss.2011.41

Cited by 25 publications

(11 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Equivalently, the power consumption upper bound is given by Eq. (53), as shown at the top of this page, stemming from the addition of Eqs (43), (45), (47) and (49). Observe in Eqs.…”

Section: A Ndqio Complexitymentioning

confidence: 99%

Non-Dominated Quantum Iterative Routing Optimization for Wireless Multihop Networks

et al. 2015

View full text Add to dashboard Cite

Routing in wireless multihop networks (WMHNs) relies on a delicate balance of diverse and often conflicting parameters, when aiming for maximizing the WMHN performance. Classified as a non-deterministic polynomial-time hard problem, routing in WMHNs requires sophisticated methods. As a benefit of observing numerous variables in parallel, quantum computing offers a promising range of algorithms for complexity reduction by exploiting the principle of quantum parallelism (QP), while achieving the optimum full-search-based performance. In fact, the so-called non-dominated quantum optimization (NDQO) algorithm has been proposed for addressing the multiobjective routing problem with the goal of achieving a near-optimal performance, while imposing a complexity of the order of O(N ) and O(N √ N ) in the best and worst case scenarios, respectively. However, as the number of nodes in the WMHN increases, the total number of routes increases exponentially, making its employment infeasible despite the complexity reduction offered. Therefore, we propose a novel optimal quantum-assisted algorithm, namely, the non-dominated quantum iterative optimization (NDQIO) algorithm, which exploits the synergy between the hardware and the QP for the sake of achieving a further complexity reduction, which is on the order of O( √ N ) and O(N √ N ) in the best and worst case scenarios, respectively. In addition, we provide simulation results for demonstrating that our NDQIO algorithm achieves an average complexity reduction of almost an order of magnitude compared with the near-optimal NDQO algorithm, while having the same order of power consumption.INDEX TERMS WMHNs, quantum computing, Pareto optimality, BBHT-QSA, DHA, NDQO.

show abstract

“…Equivalently, the power consumption upper bound is given by Eq. (53), as shown at the top of this page, stemming from the addition of Eqs (43), (45), (47) and (49). Observe in Eqs.…”

Section: A Ndqio Complexitymentioning

confidence: 99%

Non-Dominated Quantum Iterative Routing Optimization for Wireless Multihop Networks

et al. 2015

View full text Add to dashboard Cite

show abstract

“…Anytime algorithms have notably been studied for graph search [3], evaluation of belief networks [4] and GPU architectures [5].…”

Section: Related Workmentioning

confidence: 99%

“…T Q(x(t)−xref (t))+u(t) T Ru(t) (5) Note that since we have access to the true position and velocities (x(t)) of the hexrotor with the Vicon system, we can obtain the true tracking cost. Table II shows the mean of the above function over the 10 flights for both MPC across all fixed modes and RAMPC with different values of α.…”

Section: Jtrue(t) = (X(t)−xref (T))mentioning

confidence: 99%

Co-design of Anytime Computation and Robust Control

Pant

Abbas

Mohta

et al. 2015

2015 IEEE Real-Time Systems Symposium

Self Cite

View full text Add to dashboard Cite

Abstract-Control software of autonomous robots has stringent real-time requirements that must be met to achieve the control objectives. One source of variability in the performance of a control system is the execution time and accuracy of the state estimator that provides the controller with state information. This estimator is typically perception-based (e.g., Computer Vision-based) and is computationally expensive. When the computational resources of the hardware platform become overloaded, the estimation delay can compromise control performance and even stability. In this paper, we define a framework for co-designing anytime estimation and control algorithms, in a manner that accounts for implementation issues like delays and inaccuracies. We construct an anytime perception-based estimator from standard off-the-shelf Computer Vision algorithms, and show how to obtain a trade-off curve for its delay vs estimate error behavior. We use this anytime estimator in a controller that can use this tradeoff curve at runtime to achieve its control objectives at a reduced energy cost. When the estimation delay is too large for correct operation, we provide an optimal manner in which the controller can use this curve to reduce estimation delay at the cost of higher inaccuracy, all the while guaranteeing basic objectives are met. We illustrate our approach on an autonomous hexrotor and demonstrate its advantage over a system that does not exploit co-design.

show abstract

“…They also introduced a GPU kernel merging approach to merge two kernels into a single kernel for reduced sharing overheads. By using the similar kernel merging approach to reduce GPU resource underutilization, Saba et al [21] presented an algorithm that allocates GPU resources for tasks based on the resource goals and workload size. While they target a different problem of a time bound algorithm that optimizes the execution path and output quality, the employed GPU sharing approach is to merge the kernels similarly.…”

Section: Related Workmentioning

confidence: 99%

“…We use T tm and T tm op io to represent the execution time of Time Sharing under PS-1 and optimized I/O concurrency under PS-2, respectively. As T tm can be derived using previous Equation (19), T tm op io can be derived with Equation (21).…”

Section: Time Sharing: Concurrent Kernel Execution Vs Concurrent I/omentioning

confidence: 99%

Exploring Graphics Processing Unit (GPU) Resource Sharing Efficiency for High Performance Computing

Narayana

El‐Ghazawi

2013

Computers

View full text Add to dashboard Cite

The increasing incorporation of Graphics Processing Units (GPUs) as accelerators has been one of the forefront High Performance Computing (HPC) trends and provides unprecedented performance; however, the prevalent adoption of the Single-Program Multiple-Data (SPMD) programming model brings with it challenges of resource underutilization. In other words, under SPMD, every CPU needs GPU capability available to it. However, since CPUs generally outnumber GPUs, the asymmetric resource distribution gives rise to overall computing resource underutilization. In this paper, we propose to efficiently share the GPU under SPMD and formally define a series of GPU sharing scenarios. We provide performance-modeling analysis for each sharing scenario with accurate experimentation validation. With the modeling basis, we further conduct experimental studies to explore potential GPU sharing efficiency improvements from multiple perspectives. Both further theoretical and experimental GPU sharing performance analysis and results are presented. Our results not only demonstrate the significant performance gain for SPMD programs with the proposed efficient GPU sharing, but also the further improved sharing efficiency with the optimization techniques based on our accurate modeling.

show abstract

Anytime Algorithms for GPU Architectures

Cited by 25 publications

References 21 publications

Non-Dominated Quantum Iterative Routing Optimization for Wireless Multihop Networks

Non-Dominated Quantum Iterative Routing Optimization for Wireless Multihop Networks

Co-design of Anytime Computation and Robust Control

Exploring Graphics Processing Unit (GPU) Resource Sharing Efficiency for High Performance Computing

Contact Info

Product

Resources

About