Markus Frank

Analysis of Series of Measurements from Job-Centric Monitoring by Statistical Functions

The rising number of executed programs (jobs) enabled by the growing amount of available resources from Clouds, Grids, and HPC (for example) has resulted in an enormous number of jobs. Nowadays, most of the executed jobs are mainly unobserved, so unusual behavior, non-optimal resource usage, and silent faults are not systematically searched and analyzed. Job-centric monitoring enables permanent job observation and, thus, enables the analysis of monitoring data. In this paper, we show how statistic functions can be used to analyze job-centric monitoring data and how the methods compare to more-complex analysis methods. Additionally, we present the usefulness of job-centric monitoring based on practical experiences.

Challenges in Multicore Performance Predictions

Software performance predictions are an established part of an engineering like software development process and therefore relevant to enable high quality and to ensure requirement fulfillment. Software Performance Engineers use for that model-based performance predictions approaches. However, current predictions approaches are based on the assumption of single core CPU systems. To enable Software Performance Engineers to further give accurate predictions also for multicore systems, which are by now state of the art, we need to adapt our current prediction models. On the poster, we discuss the upcoming challenges to be tackled to increase the accuracy of the performance predictions models.

CAUS: An Elasticity Controller for a Containerized Microservice

Recent trends towards microservice architectures and containers as a deployment unit raise the need for novel adaptation processes to enable elasticity for containerized microservices. Microservices facing unpredictable workloads need to react fast and match the supply as closely as possible to the demand in order to guarantee quality objectives and to keep costs at a minimum. Current state-of-the-art approaches, that react on conditions which reflect the need to scale, are either slow or lack precision in supplying the demand with the adequate capacity. Therefore, we propose a novel heuristic adaptation process which enables elasticity for a particular containerized microservice. The proposed method consists of two mechanisms that complement each other. One part reacts to changes in load intensity by scaling container instances depending on their processing capability. The other mechanism manages additional containers as a buffer to handle unpredictable workload changes. We evaluate the proposed adaptation process and discuss its effectiveness and feasibility in controlling autonomously the number of replicated containers.

Time-Aligned Similarity Calculations for Job-Centric Monitoring

In job-centric monitoring, monitors gather series of measurements, e.g., the used CPU load, per job. In domains where jobs are expected to behave similar, job-centric monitoring allows detecting misbehaving jobs based on a reference series of measurements. However, current detection approaches neglect time-drifts in series, e.g., caused by different CPU speeds and therefore potentially cause false positives.To cope with this issue, this paper introduces a novel approach to compensate such time-drifts. Our approach is based on a transformation that aligns a series of measurements to the reference series time. In a proof-of-concept with synthetic job-centric monitoring data, we show that our approach reduces the number of false positives significant.