This work addresses performance testing for monitoring mass quantities of large-dataset measurements in infrastructure-as-a-Service (IaaS). Physical resources are not virtualized in sharing dynamic clouds; thus, shared resources compete for access to system resources. This competition introduces significant new challenges when assessing the performance of IaaS. A bottleneck may occur if one system resource is critical to IaaS; this may shut down the system and services, which would reduce the workflow performance by a large margin. To protect against bottlenecks, we propose CloudPT, a performance test management framework for IaaS. CloudPT has many advantages: (I) high-efficiency detection; (II) a unified end-to-end feedback loop to collaborate with cloud-ecosystems management; and (III) a troubleshooting performance test. This paper shows that CloudPT efficiently identifies and detects bottlenecks with a minimal false-positive rate (<13%) and it correlates high accuracy using the failure of a host virtual machine (host VM) to start-up with both cloud illustrative batches and transactional workloads such as the Spark, and Kafka framework for a data partitioning and collecting events on an each server. In a framework based on a trace case study, CloudPT diagnosed performance bottlenecks in 20 s with a precision rate of 86%, confirming its real-time efficiency.
CITATION STYLE
Alkasem, A., Liu, H., & Zuo, D. (2018). CloudPT: Performance testing for identifying and detecting bottlenecks in IaaS. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 11336 LNCS, pp. 432–452). Springer Verlag. https://doi.org/10.1007/978-3-030-05057-3_33
Mendeley helps you to discover research relevant for your work.