Selected Publications

This work proposes an efficient, effective and robust method to find the best configuration in the cloud. SCOUT identifies resource requirements using low-level performance metrics and searches only the spotlight region (configuration space). Our evaluation shows SCOUT is several times better than the state-of-the-art methods.
In USENIX ATC’18 (submitted), 2018.

This work identifies the fragility problem in applying Bayesian Optimization in searching for the best cloud configuration. We propose a low-level augmented Bayesian Optimization method to alleviate the fragility problem. Based on this work, we conclude that it is often insufficient to use general-purpose off-the-shelf methods for configuring cloud instances without augmenting those methods with essential systems knowledge such as CPU utilization, working memory size and I/O wait time.
In ICDCS 2018 (submitted), 2017.

Software-defined storage requires to meet users’ performance requirements. Machine learning techniques are used to create reliable performance models from low-level system metrics collected at runtime. The accurate performance model enables service providers to provision storage resources in a more fine-grained way.
In SRDS (Best Paper Award), 2016.

Recent Publications

. Scout: An Experienced Guide to Find the Best Cloud Configuration. In USENIX ATC’18 (submitted), 2018.

Preprint

. Low-Level Augmented Bayesian Optimization for Finding the Best Cloud VM. In ICDCS 2018 (submitted), 2017.

Preprint

. Trilogy: Data Placement to Improve Performance and Robustness of Cloud Computing. In SCDM, 2017.

PDF

. Inside-Out: Reliable Performance Prediction for Distributed Storage Systems in the Cloud. In SRDS (Best Paper Award), 2016.

PDF

. Evaluation of MapReduce in a large cluster. In CLOUD, 2015.

PDF

. A Novel Approach for Cooperative Overlay-Maintenance in Multi-Overlay Environments . In CloudCom, 2010.

PDF Slides

Recent Posts

Tags