Research

Operating Systems, Distributed and Real-time Systems

Overview

Operating systems provide a set of interfaces for users and programmers to achieve high utilization of hardware resources. Operating systems can be found in many computing systems, such as enterprise-level servers and embedded devices. The objectives and constraints of operating-system design are different for various types of systems. For example, security and protection could be a priority for large-scale servers; power management is a critical issue for embedded devices; timeliness is important to real-time systems.

Our faculty members focus on non-volatile memory, security for cloud systems, and parallel/peer-to-peer environment. As embedded systems, cloud computing, distributed systems rapidly developed in recent years, operating system designs also evolved quickly to support these new computing paradigms. For example, flash-based solid-state disks have been crucial components for mobile devices; attack and defense for cloud-computing systems become more and more complex and distributed; resource placement and sharing in a distributed cloud (or edge) environment is crucial; hardware-based computing becomes massively parallel. Our research members are very active in these related areas.


Prospective Views

Cloud, multicore, and non-volatile: these three factors picture the design of next-generation operating systems. In the near future, portable computers would have the extremely high computing power and get rid of conventional hard drives. Computing and application services are available in the cloud, and resource sharing would be transparent to users via peer-to-peer sharing mechanism.


Important Research Results

  • Hybrid solid-state disks(IEEE Trans. Computers)
  • Wear-leveling algorithms(ACM Trans. Design Automation for Electronic Systems)
  • SSD write buffering(Design Automation Conference)
  • xSENSE Introspection Technology:a cloud virus-scanning system。
  • Massive data analysis based on video-processors
  • Publish/subscript P2P systems


Important Research Achievements

  • Hybrid Solid-State Disks
    This hybrid design of SSDs combines high-speed SLC flash and large-capacity MLC flash for a better balance between good write performance and low device cost. This technology involves hot-data identification, cold-data migration, and flash-wear throttling. The result has been nominated as a best-paper candidate in ASP-DAC 2008 and then published in a prestigious journal IEEE Transactions on Computer.
  • SSD buffering algorithm
    The performance of SSD random write is inherently low because flash management requires address translation and garbage collection. This research aims at using a buffer memory to not only reduce write traffic to flash memory but also relieve SSDs of heavy garbage-collection overheads. The research result has been published in Design Automation Conference.
  • Flash Wear Leveling
    Lifetime has long been a critical issue for flash storage devices. This research aims at a wear-leveling algorithm for evenly distributing erase operations in flash blocks while reducing the extra data movements. The results have been published in ACM Transactions on Design Automation for Electronic Systems and had been transferred to iTE Co. Ltd.
  • The real-time virus scanning system
    xSENSE Introspection Technology for cloud-computing systems. xSENSE uses ClamAV for virus-signature recognition. xSNES captures file-accessing events in DomU virtual machines. When client systems are about to access files, xSENSE will read the virtual-disk image for the accessed files, and check virus signatures using ClamAV. The file access will be granted if the accessed file is virus-free. There are three key technologies inside of xSNESE: intercepting system calls from guest Oss, inspecting memory of guest Oss, and accessing files in virtual-disk images.
  • Massive-data analysis based on video-processor acceleration
    This work uses a large number of computing units provided in the GPU for speeding up massive data analysis. The research result includes an in-memory database and a highly efficient query system based on GPU-acceleration.
  • Publish/Subscription P2P Systems
    This work focuses on publishing and subscription of P2P services using a common authentication mechanism called “OpenID”. With this technology, users can use P2P services without first registering to these services.


 Research Projects

Project Title

Affiliation

PI

Time

The research on IaaS based CDN service

National Science Council

Prof. Shyan-Ming Yuan

2012-2014

Design and Implementation of a Learning Assistance System for Supporting Information-Problem Solving

National Science Council

Prof. Shyan-Ming Yuan

2010-2013

Cross-layer collaborative performance optimization

strategies for solid-state disks

National Science Council

Prof. Li-Pin Chang

2012-2015

Data management and access policies for high-performance solid-state disks

National Science Council

Prof. Li-Pin Chang

2009-2012

Xen Hypervisor based Intrusion Detection and Response for IaaS Cloud

National Science Council

Prof. Yu-Sung Wu

2010-2012

Respond to Zero-day Attacks in Distributed Computing and Cloud Computing Environments

National Science Council

Prof. Yu-Sung Wu

2010-2012



Research Themes

  • Communication Middleware
  • Distributed Algorithms
  • Internet Game Platform
  • Network Simulator and Emulator
  • Service-Oriented Computing
  • Embedded Storage System


Research Team