Yong-Yeon Jo

Collaborative processing of data-intensive algorithms with CPU, intelligent SSD, and GPU

The graphic processing unit (GPU) is a computing resource to process graphics-related applications. The intelligent SSD (iSSD) is a solid state device (SSD) that is provided with data processing power. These days, CPU, GPU, and SSD are equipped together in most processing environment. If SSD is replaced with iSSD later on, we have a new processing environment where three computing resources collaborate one another to process a huge volume of data (so called big data) quite effectively. In this paper, we address how to exploit all these computing resources for efficient processing of data-intensive algorithms.Through extensive experiment, we verify the effectiveness and potential of the proposed collaborative processing environment by processing data concurrently with multiple computing resources. The results reveal that processing in the our environment outperforms that in the traditional one by up to 3.5 times.

Efficient Sparse Matrix Multiplication on GPU for Large Social Network Analysis

As a number of social network services appear online recently, there have been many attempts to analyze social networks for extracting valuable information. Most existing methods first represent a social network as a quite sparse adjacency matrix, and then analyze it through matrix operations such as matrix multiplication. Due to the large scale and high complexity, efficient processing multiplications is an important issue in social network analysis. In this paper, we propose a GPU-based method for efficient sparse matrix multiplication through the parallel computing paradigm. The proposed method aims at balancing the amount of workload both at fine- and coarse-grained levels for maximizing the degree of parallelism in GPU. Through extensive experiments using synthetic and real-world datasets, we show that the proposed method outperforms previous methods by up to three orders-of-magnitude.

On running data-intensive algorithms with intelligent SSD and host CPU: a collaborative approach

A solid state device (SSD), which has the characteristics such as high IO bandwidth and low access latency, is drawing attention as a next-generation storage device. Even though SSD provides a high internal bandwidth, the performance bottleneck exists on the host interface of relatively low bandwidth in spite of the increased internal bandwidth of SSD. To overcome the performance bottleneck, the notion of intelligent SSD (iSSD) has been proposed. In iSSD, there are still problems in processing the algorithms of high complexity. In this paper, we address an effective collaboration of iSSD and host CPU in order to maximize the performance of data-intensive algorithms. Extensive experimental results show that our approach performs faster up to 2.43 times than a previous approach.

Intelligent SSD: A turbo for big data mining

This research was supported by (1) Semiconductor Industry Collaborative Project between Hanyang University and Samsung Electronics Co. Ltd., (2) the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIP) (No. NRF-2014R1A2A1A10054151), (3) the ICT R&D program of MSIP/IITP (B0101-15-0266, Development of High Performance Visual Big-Data Discovery Platform for Large-Scale Realtime Data Analysis), and (4) the National Research Foundation of Korea (NRF) Grant funded by the Korean Government (MSIP) (No. 2015R1A5A7037751).

A High-Performance Graph Engine for Efficient Social Network Analysis.

Existing single-machine based graph engines do not leverage the characteristic of social networks following the power-law degree distribution. We propose a new graph engine tailored for processing and analyzing large-scale social networks efficiently by exploiting the power-law degree property

High-performance data mining with intelligent SSD

An intuitive way to process the big data efficiently is to reduce the volume of data transferred over the storage interface to a host system. This is the reason that the notion of intelligent SSD (iSSD) was proposed to give processing power to SSD. There is rich literature on iSSD, however, its real implementation has not been provided to the public yet. Most prior work aims to quantify the benefits of iSSD with analytical modeling. In this paper, we first develop on iSSD simulator and present the potential of iSSD in data mining through the iSSD simulator. Our iSSD simulator performs on top of the gem 5 simulator and fully simulates all the processes of data mining algorithms running in iSSD with cycle-level accuracy. Then, we further addresse how to exploit all the computing resources for efficient processing of data mining algorithms. These days, CPU, GPU, and SSD are recently equipped together in most computing environment. If SSD is replaced with iSSD later on, we have a new computing environment where the three computing resources collaborate one another to process big data quite effectively. For this, scheduling is required to decide which computing resource is going to run for which function at which time. In our heterogeneous scheduling, types of computing resources, memory sizes in computing resources, and inter-processor communication times including IO time in SSD are considered. Our scheduling results show that processing in the collaborative environment outperforms that in the traditional one by up to about 10 times.

Data mining in intelligent SSD: Simulation-based evaluation

Due to an explosive growth of Internet applications, the amount of data has increased enormously. In order to store and process this big data more efficiently, a solid-state device (SSD) has replaced a hard disk drive (HDD) as a primary storage media. In spite of high internal bandwidth, SSD has its performance bottleneck on the host interface whose bandwidth is relatively low. To overcome the problem of performance bottleneck in big data processing, the notion of intelligent SSD (iSSD) was proposed to give computing power to SSD. However, its real implementation has not been provided to the public yet. In this paper, we are going to verify the potential of iSSD in handling data-intensive algorithms. To the end, we first develop an iSSD simulator and then evaluate the performance of data mining algorithms inside iSSD on the top of it in comparison with that by the host CPU. The results reveal that data mining with iSSD outperforms that with host CPUs up to around 300%.

Data Locality in Graph Engines: Implications and Preliminary Experimental Results

The size of graphs has dramatically increased. Graph engines for a single machine have been emerged to process these graphs efficiently. However, existing engines have overlooked a data locality which is an imperative factor to improve the performance of these engines in the previous literature. In this paper, we show the importance of data locality with graph algorithms by running on graph engines based on a single machine.

Efficient computations of link-based similarity measures on the GPU

In this paper, we first analyze how the characteristics of the GPU affect the performance of link-based similarity measures. Based on the analysis, we describe our strategies to improve the performance of link-based similarity measures on the GPU in detail. Finally, through extensive experiments, we evaluate the effectiveness of our strategies.

Efficient processing of recommendation algorithms on a single-machine-based graph engine

The wide use of recommendation systems includes more users and items in system operations, leading to a significant increase in the size of related datasets. However, recommendation algorithms on existing single-machine-based graph engines have been developed without considering the important characteristics of recommendation datasets, i.e., huge size and power-law degree distribution. In this paper, we address how to realize efficient graph- and matrix-factorization-based recommendation algorithms, handling recommendation datasets on RealGraph, a state-of-the-art single-machine-based graph engine. Through extensive experiments, we demonstrate that our recommendation algorithms on RealGraph universally and consistently outperform the algorithms on other graph engines over all datasets up to 34 times.