A practical method for estimating performance degradation on multicore processors, and its application to HPC workloads

Abstract

Abstract-When multiple threads or processes run on a multicore CPU they compete for shared resources, such as caches and memory controllers, and can suffer performance degradation as high as 200%. We design and evaluate a new machine learning model that estimates this degradation online, on previously unseen workloads, and without perturbing the execution.Our motivation is to help data center and HPC cluster operators effectively use workload consolidation. Data center consolidation is about placing many applications on the same server to maximize hardware utilization. In HPC clusters, processes of the same distributed applications run on the same machine. Consolidation improves hardware utilization, but may sacrifice performance as processes compete for resources. Our model helps determine when consolidation is overly harmful to performance. Our work is the first to apply machine learning to this problem domain, and we report on our experience reaping the advantages of machine learning while navigating around its limitations. We demonstrate how the model can be used to improve performance fidelity and save energy for HPC workloads.

Enabling Fair Pricing on High Performance Computer Systems with Node Sharing

“…There are also studies that evaluate the effectiveness of analytical and statistical models to solve problems related to contention [16,25,36,63]. The computational complexity, heuristics and approximation algorithms for optimal multiprocessor scheduling are explored in [10,23,30,32].…”

Characterization and modeling of PIDX parallel I/O for performance optimization

“…Moreover, unlike most machine learning models, tree-based models do not require much parameter tuning. Indeed, tree-based models such as random forests and GBDT are a popular choice for model building and data analytics and have proved useful in related HPC applications [9]. In the discussions below, we present our results using tree-based models.…”

“…Table 4 also presents an example line from our combined dataset. Attribute selection or attribute reduction procedures [9] to select a smaller set of useful attributes are usually beneficial for large attribute spaces. In our case, since the number of attributes was already small and moderately reasonable, we did not perform any attribute selection or reduction.…”

“…Earlier work in high-performance research has proposed analytical models, heuristic models, and trial-and-error based approaches. All these methods have known limitations [9] and do not generalize well to a wide variety of settings. Modeling techniques based on machine learning overcome these system limitations and build a knowledge-based model that is independent of the specific hardware, underlying file system, or custom library used.…”

Enabling fair pricing on HPC systems with node sharing

“…There have also been studies that evaluate the effectiveness of analytical and statistical models to solve problems related to contention [40,60,24,31]. The computational complexity, heuristics and approximation algorithms for optimal multiprocessor scheduling are explored in [30,19,37,35].…”