I am the last year PhD candidate of Computer Science at the University of Virginia. I am working with Prof. Marty Humphrey -- he is a truly awesome advisor (). My research area is Cloud Computing and Distributed Systems with emphasis on building effective resource management systems and evaluation environments for user applications on the clouds. Current research direction is focused on performance modeling, workload prediction, and simulation-based evaluation for cloud resource management system and its applications.

Before coming to UVa, I was a software research engineer at LG-Ericsson. I was in charge of research and development of cluster management middleware for large-scale telecommunication systems.

For more information, please refer to my CV.

Research Projects

Performance Study of Workload Predictors for Predictive Cloud Resource Management - CLOUD '16

Many predictive resource scaling approaches have been proposed to overcome the limitations of the conventional reactive approaches most often used in clouds today. In general, due to the complexity of clouds, these predictive approaches were often forced to make significant limiting assumptions in either the operating conditions/requirements or expected workload patterns. As such, it is extremely difficult for cloud users to know which – if any – existing workload predictor will work best for their particular cloud activity, especially when considering highly-variable workload patterns, non-trivial billing models, a variety of resources to add/subtract, etc.

To solve this problem, we conduct comprehensive evaluations for a variety of workload predictors. The workload predictors cover four classes of 21 predictors:
A) Naive: mean-based and kNN
B) Global and local regression: linear and polynomial regression
C) Time-series models: smoothing-based (e.g. ES and WMA) and Box-Jenkins models (e.g. AR, ARMA, and ARIMA)
D) machine learning-based methods: SVM, Decision Tree, and Ensemble Approaches

We created a simulation model for cloud applications with predictive scaling components on top of PICS. We measured the performance of the various predictors and predictive scaling models with various/realistic workload patterns, two different cloud billing models, and three different styles of predictive scaling.

Our evaluation confirms that no workload predictor is universally best for all workload patterns, and shows that Predictive Scaling-out + Predictive Scaling-in approach has the best cost efficiency and the lowest job deadline miss rate in cloud resource management, on average providing 30% better cost efficiency and 80% less job deadline miss rate compared to other styles of predictive scaling.

PICS: A Public IaaS Cloud Simulator [Project Page] - CLOUD '15

Public clouds become essential for many organizations to run their applications because clouds offer huge financial benefits and great flexibility (e.g. pay-as-you-go ☺). However, it is very challenging to accurately evaluate the performance and cost of user applications without actual deployment on the clouds. Existing cloud simulators are generally designed from the perspective of cloud service providers, thus they can be under-developed for answering questions for the perspective of cloud users.

To solve this prediction and evaluation problem, we created PICS -- a Public Cloud IaaS Simulator. PICS enables the cloud user to evaluate the cost and performance of public IaaS clouds along with such dimensions like VM and storage service, resource scaling, job scheduling, and diverse workload patterns. We extensively validated PICS by comparing its results with the data acquired from real public IaaS cloud using real cloud-applications. We show that PICS provides highly accurate simulation results (less than 5% of average errors) under a variety of use cases. Moreover, we evaluated PICS’ sensitivity with imprecise simulation parameters. The results show that PICS still provides very reliable simulation results with imprecise simulation parameters and performance uncertainty.

Read More and Enjoy PICS!

WDCloud: An End to End System for Large-Scale Watershed Delineation on Cloud - BigData in Geoscience '15

Watershed delineation is a process to compute the drainage area for a point on the land surface, which is a critical step in hydrologic and water resources analysis. However, existing watershed delineation tools are still insufficient to support hydrologists and watershed researchers due to the lack of essential capabilities such as fully leveraging scalable and high-performance computing infrastructure (public cloud -- e.g. Amazon Web Services, Microsoft Azure, and Google Compute Engine), and providing predictable performance for the delineation tasks.

To solve these problems, this paper reports on WDCloud: a system for large-scale watershed delineation on the public cloud. For the design and implementation of WDCloud, we employ three main approaches:
1) an automated catchment search mechanism for a public data set. i.e. NHD+
2) three performance improvement strategies (Data-reuse, parallel-union, and MapReduce).
3) machine learning-based execution time estimator for watershed delineation.

Moreover, WDCloud extensively utilizes several compute and storage capabilities from Amazon Web Services in order to maximize the performance, scalability, and elasticity of watershed delineation system. Our evaluations on WDCloud focus on two main aspects of WDCloud; the performance improvement for watershed delineation via three strategies and the estimation accuracy for watershed delineation time by local linear regression. The evaluation results show that WDCloud can achieve 18x–111x of speed-ups for delineating any scale of watersheds in the contiguous United States as compared to commodity laptop environments, and accurately predict execution time for watershed delineation with 85.6% of prediction accuracy, which is 23%–43% higher than other state-of-the-art approaches.

LCA: Elastic Resource Managemet System for Scientific Applications on Public Clouds - UCC '14

Scientists have become increasingly reliant on large-scale compute resources on public IaaS clouds to efficiently process their applications. Unfortunately, the reactive nature of autoscaling techniques made available by the public cloud provider can cause insufficient response time and poor job deadline satisfaction rates.

To solve these problems, we designed LCA: an end-to-end elastic resource management system for scientific applications on public IaaS clouds. This system employs three following "L" - "C" - "A" strategies:
· Local regression-based "accurate" job execution time predictor.
· Cost/performance balanced resource evaluation/selection approach.
· Availability-aware job scheduling algorithm.

This comprehensive system is deployed on Amazon Web Services and is compared with other state-of-the-art resource management schemes. Experimental results show that our system achieves a 9%–32% improvement with respect to the deadline satisfaction rate over other schemes. We achieve this deadline satisfaction rate improvement while still providing improved cost-efficiency over other state-of-the-art approaches.


iCSI: A Cloud Garbage VM Collector for Addressing Inactive VMs with Machine Learning
In Kee Kim, Sai Zeng, Christopher Young, Jinho Hwang and Marty Humphrey
Proceedings of 2017 IEEE International Conference on Cloud Engineering (IC2E 2017)
[paper] [slide] [poster] [link]

A Supervised Learning Model for Identifying Inactive VMs in Private Cloud Data Centers
In Kee Kim, Sai Zeng, Christopher Young, Jinho Hwang and Marty Humphrey
Proceedings of 2016 ACM/IFIP/USENIX International Middleware Conference, Industry Track (Middleware 2016)
[paper] [slide] [link]

Empirical Evaluation of Workload Forecasting Techniques for Predictive Cloud Resource Scaling
In Kee Kim, Wei Wang, Yanjun Qi and Marty Humphrey
Proceedings of 2016 IEEE International Conference on Cloud Computing (IEEE CLOUD 2016)
[paper] [slide

WDCloud: An End to End System for Large-Scale Watershed Delineation on Cloud
In Kee Kim, Jacob Steele, Anthony Castronova, Jonathan Goodall and Marty Humphrey
Proceedings of 2015 IEEE International Conference on Big Data (IEEE BigData 2015) Workshop (GeoBigData '15)
[paper] [slide] [link]

PICS: A Public IaaS Cloud Simulator
In Kee Kim, Wei Wang, and Marty Humphrey
Proceedings of 2015 IEEE International Conference on Cloud Computing (IEEE CLOUD 2015)
[paper] [project page] [slide] [link]

Toward Optimal Resource Provisioning for Cloud MapReduce and Hybrid Cloud Applications
Arkaitz Ruiz-Alvarez, In Kee Kim, and Marty Humphrey
Proceedings of 2015 IEEE International Conference on Cloud Computing (IEEE CLOUD 2015)

Comprehensive Elastic Resource Management to Ensure Predictable Performance for Scientific Applications on Public IaaS Clouds
In Kee Kim, Jacob Steele, Yanjun Qi, Marty Humphrey
Proceedings of 2014 IEEE/ACM International Conference on Utility and Cloud Computing (IEEE/ACM UCC 2014)
[paper] [slide] [link]

CloudDRN: A Lightweight, End-to-End System for Sharing Distributed Research Data in the Cloud
Marty Humphrey, Jacob Steele, In Kee Kim, Michael Kahn, Jessica Bondy, Micheal Ames
Proceedings of 2013 IEEE International Conference on e-Science (IEEE eScience 2013)

Adaptive and Mobility-Predictive Quantization-based Communication Data Management in High-Performance Distributed Computing
In Kee Kim, Sung Ho Jang and Jong Sik Lee
Simulation: Transactions of The Society for Modeling and Simulation International Vol. 83, No. 7 2007
[paper] [link]

Adaptive Distance Filter-based Traffic Reduction for Mobile Grid
In Kee Kim, Sung Ho Jang and Jong Sik Lee
Proceedings of 27th International Conference on Distributed Computing Systems (ICDCS 2007) Workshop
[pdf] [link]
Ref 7 and 8 are papers from my current adviser (Prof. Marty Humphrey )

Resource Demand Prediction-based Grid Resource Transaction Network Model in Grid Computing Environment
In Kee Kim and Jong Sik Lee
Proceedings of 6th International Conference on Computational Science and Its Applications (ICCSA 2006)