Myungcheol Lee

GPU accelerated item-based collaborative filtering for big-data applications

Recommendation systems are a popular marketing strategy for online service providers. These systems predict a customers future preferences from the past behaviors of that customer and the other customers. Most of the popular online stores process millions of transactions per day; therefore, providing quick and quality recommendations using the large amount of data collected from past transactions can be challenging. Parallel processing power of GPUs can be used to accelerate the recommendation process. However, the amount of memory available on a GPU card is limited; thus, a number of passes may be required to completely process a large-scale dataset. This paper proposes two parallel, item-based recommendation algorithms implemented using the CUDA platform. Considering the high sparsity of the user-item data, we utilize two compression techniques to reduce the required number of passes and increase the speedup. The experimental results on synthetic and real-world datasets show that our algorithms outperform the respective CPU implementations and also the naïve GPU implementation which does not use compression.

Validation of XML Document Updates Based on XML Schema in XML Databases

We study the validation of XML documents when they are updated in XML databases. An XML document can be verified by checking against an XML Schema, which contains structure and type information of XML documents. However, most of XML database systems just validate the whole XML document, but can not validate parts of it. If updates are very frequent, then validating the whole XML document will cause serious performance degradation. Furthermore, rollback should be performed if the updates result in an invalid document, because the updated document is usually validated after the update operation executed. In this paper, we propose an immediate and partial validation mechanism for solving these two problems, i.e the validity of an update operation is checked immediately before the actual update is applied to the database whether it causes invalidity, and validation is performed only on the updated parts of the XML document in the database. Consequently, XML database systems can maintain valid XML documents at any time. We already proposed an immediate and partial validation mechanism based on DTD[6], and we extend the mechanism based on XML Schema in this paper.

Load adaptive and fault tolerant distributed stream processing system for explosive stream data

As smart devices such as sensors, smartphones, and CCTVs are becoming extensively utilized recently, stream data from those smart devices are consistently generated explosively. There are also increasing cases that we notice security attacks after already important assets are damaged by cyber-targeted attacks such as APT attacks due to the lack of real-time security log processing capability. Accordingly, the demand to process and analyse the exploding stream data in real-time and in advance is consistently increasing in many application domains. However, existing distributed stream processing systems like Storm and S4 are not well adaptive when there are drastic increase of input stream data. In this paper, we propose a distributed stream processing system which supports several load adaptation techniques utilizable for various circumstances of explosive data stream, and also supports fault tolerance mechanisms to fail over in several failure situations.

Load adaptive distributed stream processing system for explosive stream data

As smart devices such as sensors, smartphones, and CCTVs are becoming extensively utilized recently, stream data from those smart devices are consistently generated explosively. There are also increasing cases that we notice security attacks after already important assets are damaged by cyber-targeted attacks such as APT attacks due to the lack of real-time security log processing capability. Accordingly, the demand to process and analyse the exploding stream data in real-time and in advance is consistently increasing in many application domains. However, existing distributed stream processing systems like Storm and S4 are not well adaptive when there are drastic increase of input stream data. In this paper, we propose a distributed stream processing system which supports several load adaptation techniques utilizable for various circumstances of explosive data stream.

A JIT Compilation-Based Unified SQL Query Optimization System

In-memory databases are gaining attention as a solution to efficiently support SQL queries on large volume of data, as main memories are becoming cheaper and grow in size. However, their query performance is not well improved on modern hardware with faster CPUs, registers and caches due to the limitation of the classical iterator style query processing model. We propose a unified SQL query optimization system using JIT compilation of OLTP, OLAP, and Stored Procedure workloads for enhanced performance on modern hardware.

Design of a scalable data stream channel for big data processing

This paper outlines big data infrastructure for processing data streams. Our project is distributed stream computing platform that provides cost-effective and large-scale big data services by developing data stream management system. This research contributes to advancing feasibility of big data processing for distributed, real-time computation even when they are overloaded.

Real-Time Data Flow Language Processing System for Handling Streams of Data

Apache Pig system generates MapReduce jobs by compiling program scripts written in Pig Latin to process large data sets in parallel on distributed computing nodes. There are inefficient features in Pig due to the limitation of the MapReduce, e.g., the MapReduce is used only for batch processing. As various smart devices are extensively utilized recently, streams of data are generated explosively and the need to process streams of data in real-time is required. In this paper, we propose a data flow language processing system, called LAMA-CEP, by generating DAG-based stream processing services to process unbounded streams of data in real-time continuously. We present a stream processing language, called Pig Latin Stream extended from Pig Latin. Programs written in Pig Latin Stream are translated into distributed stream processing jobs and then the jobs are executed on a highly scalable distributed stream processing system to process large streams of data in real-time.

A cursor-based approach to access of integrated results in mediator systems

In information integration systems based on mediator/wrapper architecture, returning all the integrated results to the mediator client at once could incur great disaster on the mediator client. If the query is not specified precisely by the user, the server would generate too many irrelevant results. And users of the Web have a tendency not to access all the results though the results are relevant to them. Moreover, when the mediator clients are deployed on mobile devices with limited hardware capacity and network resources, the situation becomes worse. In this paper, we propose a new mechanism for transferring integrated results partially to mediator clients by adopting the cursor notion of relational DBMS.