Woo Hyun Ahn

DFS: a de-fragmented file system

Small file accesses are still limited by disk head movement on modern disk drives with the high disk bandwidth. Small file performance can be improved by grouping and clustering, each of which places multiple files in a directory and places blocks of the same file on disks contiguously. These schemes make it possible for file systems to use large data transfers in accessing small files, reducing disk accesses. However, as file systems become aged, disks become too fragmented to support the grouping and clustering of small files. This fragmentation makes it difficult for file systems to take advantage of large data transfers, increasing disk I/Os. To offer a solution to this problem, we describe a de-fragmented file system (DFS). By using data cached in memory, DFS relocates and clusters data blocks of small fragmented files in a dynamic manner. Besides, DFS clusters related small files in the same directory, at contiguous disk locations. Measurements of DFS implementation show that the techniques alleviate file fragmentation significantly and, in particular performance for small file reads exceeds that of a traditional file system by 78%.

Efficient cooperative caching for file systems in cluster-based Web servers

Most of the traditional cooperative caching schemes were developed so that remote users may share files in network file systems, but were not designed for supporting recent cluster based Web servers with content based request distribution. For this reason, although the schemes are applied to the Web servers, their performance suffers from high disk accesses and large block access latency. The paper proposes and evaluates a new cooperative caching suitable for file systems in the Web servers. By exploiting the characteristics of the Web servers, it performs a cache replacement, called Duplicate First copy Replacement (DFR) to avoid caching unnecessary data produced during serving remote client requests, and thus minimizes disk accesses. Also, it reduces the overhead and block access latency required to fetch a file block (or page) in the cooperative cache with regard to both the content based Web server architecture and characteristics. Trace-driven simulation shows that our cooperating caching decreases the disk access ratio by 29%, and reduces block access latency by about 26% more than the existing cooperative algorithms.

Applications IO profiling and analysis for smart devices

Nowadays, Android has been widely installed in mobile devices, including smart-phones, tablet PCs, and PMPs, that are equipped with a wealth of standard capabilities, such as touchpad display, wireless communications, GPS, and memory-based storage. In Android systems, most users perform many activities not only phone calls, but also internet browsing, emailing, playing games, taking and sharing pictures, downloading and reading mobile e-books, and using multimedia applications. These activities and user responsiveness are mainly affected by Android IO subsystems since the Android framework uses many IO operations with framework components. In Android-based systems, IO subsystems are organized with both internal system storage and external data storage. We analyze storage traffic of three android killer applications, Gallery, Web browsing, and SNS services, by profiling storage IO for various user activities. Based on this analysis, we summarize the statistical pattern of various Android applications, and show the effect of the proposed IO write buffering scheme which reduces the number of writes for the storage system and increases latency.

Content-aware cooperative caching for cluster-based web servers

Most traditional cooperative caching schemes were developed for network file systems but were not designed for cluster-based web servers with content-based request distribution. Hence, although the schemes are applied to the web servers, their performance is not good due to high disk accesses and large block access latency. This paper proposes and evaluates a new cooperative caching scheme suitable to file systems in web servers. It uses a cache replacement policy, called duplicate first copy replacement, to avoid caching unnecessary data produced during serving requests and to minimize disk accesses. In addition it reduces block access latency required to fetch a file-block in the cooperative cache. A simulation shows that our cooperating caching decreases the disk access ratio by 29%, and reduces block access latency by about 26% when compared to the existing cooperative algorithms.

Reengineering Template-Based Web Applications to Single Page AJAX Applications

Web pages in a template-based web application (TWA) are automatically populated using a template shared by the pages with contents specific to the pages. So users can easily obtain information guided by a consistent structure of the template. Reduced duplicated code helps to increase the level of maintainability as well. However, TWA still has the interaction problem of classic web applications that each time a user clicks a hyperlink a new page is loaded, although a partial update of the page is desirable. This paper proposes a reengineering technique to transform the multi-page structure of legacy Java-based TWA to a single page one with partial page refresh. In this approach, hyperlinks in HTML code are refactored to AJAX-enabled event handlers to achieve the single page structure. In addition, JSP and Servlet code is transformed in order not to send data unnecessary for the partial update. The new single page consists of individual components that are updateable independently when interacting with a user. Therefore, our approach can improve interactivity and responsiveness towards a user while reducing CPU and network usage. The measurement of our technique applied to a typical TWA shows that our technique improves the response time of user requests over the TWA in the range from 1 to 87%.

A server framework for scheduling multimedia applications in open system environment

We propose methods for scheduling multimedia soft real-time tasks in an open system environment. We discuss two issues, guaranteeing the schedulability of every task in the application and meeting the QoS of multimedia soft real-time tasks. First, when a real-time application is composed of a heterogeneous task set, hard real-time tasks and multimedia soft real-time tasks, the operating system must guarantee the schedulability of each task in the application. These guarantees can be achieved by preserving the CPU bandwidth of each task in the application. We have designed a server framework that preserves the CPU bandwidth of each task in the application. Each task can select its own server that can satisfy its time constraints. Second, when an operating system schedules multimedia soft real-time tasks in an open system, it is important for the operating system not only to support sufficient QoS level for the task, but also to utilize the CPU bandwidth of the system efficiently. We invented a new method of utilizing the CPU bandwidth of the system efficiently while supporting the requested QoS level.

Bitmap discard operation for the higher utilization of flash memory storage

NAND Flash memory storage is widely used in computing systems. In General, there exists mismatch between logical address and physical address in Flash storage, and these address translations are managed by Flash Translation Layer (FTL). Due to its management, logically invalid data is considered as physically valid at some parts in Flash device, which causes additional overhead. The physically valid area being logically invalid area can be invalidated by TRIM or discard command, however, too many discard commands degrade throughput. In this paper, we propose a bitmapbased discard operation which can decrease the number of runtime discard commands. According to the proposed scheme, hundreds of separated region can be discarded all at one bitmap-based discard command.