Job placement advisor based on turnaround predictions for HPC hybrid clouds

Abstract

Abstract-Several companies and research institutes are moving their CPU-intensive applications to hybrid High Performance Computing (HPC) cloud environments. Such a shift depends on the creation of software systems that help users decide where a job should be placed considering execution time and queue wait time to access on-premise clusters. Relying blindly on turnaround prediction techniques will affect negatively response times inside HPC cloud environments. This paper introduces a tool to make job placement decisions in HPC hybrid cloud environments taking into account the inaccuracy of execution and waiting time predictions. We used job traces from real supercomputing centers to run our experiments, and compared the performance between environments using real speedup curves. We also extended a stateof-the-art machine learning based predictor to work with data from the cluster scheduler. Our main findings are: (i) depending on workload characteristics, there is a turning point where predictions should be disregarded in favor of a more conservative decision to minimize job turnaround times and (ii) scheduler data plays a key role in improving predictions generated with machine learning using job trace data-our experiments showed around 20% prediction accuracy improvements.

Helping HPC Users Specify Job Memory Requirements via Machine Learning

“…This method is used in production in the XSEDE grid to predict queue waiting time. We have implemented this method ourselves for hybrid cloud environments with reasonably good results for both running time and queue waiting time [8]. However, as one can see from Tables IV and V other methods outperformed this form of kNN in several segments.…”

Predictive performance modeling for distributed batch processing using black box monitoring and machine learning

“…Site selection was intensively studied in grid computing, motivated by the availability of various clusters for a given job [46]. Another application area are systems that chose the best from a set of available clouds [47]. Clearly, the expected runtime of a job on a site (or cloud) is an important parameter in any such decision method.…”

“…Cunha et al [47] consider performance differences between a local cluster and cloud resources. They assume that executing in the cloud slows down execution by a constant, applicationdependent factor, for which they propose a linear model and an empirical model, based on relative performance of eight applications on three clusters and on three cloud platforms.…”

VM reassignment in hybrid clouds for large decentralised companies: A multi-objective challenge

“…Examining the available literature in hybrid clouds, we see that the majority of these studies focus on architecture / management aspects of such environments and their challenges [25]. Studies concerned with resource allocation are often either (i) focused on the selection of the public cloud by formalising it as a brokering problem (e.g., with multiple cloud providers [49] or with a distributed cloud [20]) or (ii) focused on the private DC with the possibility of bursting to public clouds [15] to cope with a momentary excess of demand in resources.…”

“…Works that deal with VM allocation in hybrid data centres are often mono-objective (e.g., driven by overall cost reduction [3], SLA satisfaction [15] or minimisation of VM leasing cost [10]). In addition, they either (i) address a small scale scheduling problem with a simplistic private DC architecture: Chunlin and LaYuan [13] define a set of routines for an optimal resolution of the problem, while Bittencourt et al [10] surveys common scheduling algorithms and studies their performance on the problem, or (ii) put an emphasis on the private DC sub-problem [12].…”