prediction of its computational complexity. This is addressed, in this paper, by using several neural network modules, each of which is appropriate for a given rendering process. For this reason, initially, a feature vector is constructed to describe with high efficiency the parameters affected

{lohmann, wosch} @ cs.fau.de The trend towards many-core systems comes with various is-sues, among them their highly dynamic and non-predictable workloads. Hence, new paradigms for managing resources of many-core systems are of paramount importance. The problem of resource management, e.g. mapping

the effect of drivers ’ responses to unacceptable v&i & states that build up while not fully attending, an entropy measure is used as an index of behavioral predictability and used 1 ” characterize the workload induced by different tasks. Workload is considered high over a given time interval when

as models to predict the combined resource utilization of those workloads. We formalize the consolidation problem as a non-linear optimization program, aiming to minimize the number of servers and balance load, while achieving near-zero performance degradation. We compare Kairos against virtual machines

. Neural network performs workload prediction by modeling the non-linear input-output relationship between rendering descriptors and the respective computational complexity. To increase the prediction accuracy, a constructive algorithm is adopted in this paper to train the neural network so that network

frame-work that attempts to predict the performance of similar classes of applications on different processors. Most signifi-cantly, this study identifies several non-intuitive relation-ships between program characteristics and demonstrates that it is possible to accurately model CUDA kernel perfor

predict renewable energy as well as IT demand and design an IT workload management plan that schedules IT workload and allocates IT resources within a data center according to time varying power supply and cooling efficiency. We have implemented and evaluated our approach using traces from real data

savings. Traditionally reshaping of the access workload is a result of caches passively modifying the workload with the aim of increasing hit ratios and reducing access latency. In contrast, we present the a shifting predictive policy that actively reshapes the workload with the primary goal of conserving

and their interactions. Multivariate adaptive regression splines effectively capture non-linear and non-monotonic as-sociations between the response and predictor variables. Our models are interpretable, easy to construct and exhibit high correlation/low errors between predicted and measured performance. Furthermore

(CSP) for predicting subsequent readers. We evaluate this class of predictors for the first time on commercial applications and demonstrate that our DGP correctly predicts 47%-76 % of last stores. Memory sharing patterns in commercial workloads are inherently non-repetitive; hence CSP cannot attain