Inhyeok Kim

VMMB: Virtual Machine Memory Balancing for Unmodified Operating Systems

Virtualization technology has been widely adopted in Internet hosting centers and cloud-based computing services, since it reduces the total cost of ownership by sharing hardware resources among virtual machines (VMs). In a virtualized system, a virtual machine monitor (VMM) is responsible for allocating physical resources such as CPU and memory to individual VMs. Whereas CPU and I/O devices can be shared among VMs in a time sharing manner, main memory is not amendable to such multiplexing. Moreover, it is often the primary bottleneck in achieving higher degrees of consolidation. In this paper, we present VMMB (Virtual Machine Memory Balancer), a novel mechanism to dynamically monitor the memory demand and periodically re-balance the memory among the VMs. VMMB accurately measures the memory demand with low overhead and effectively allocates memory based on the memory demand and the QoS requirement of each VM. It is applicable even to guest OS whose source code is not available, since VMMB does not require modifying guest kernel. We implemented our mechanism on Linux and experimented on synthetic and realistic workloads. Our experiments show that VMMB can improve performance of VMs that suffers from insufficient memory allocation by up to 3.6 times with low performance overhead (below 1%) for monitoring memory demand.

NEMOSHELL Demo: Windowing System for Concurrent Applications on Multi-user Interactive Surfaces

Recently, the prevalence of large interactive surfaces renewed interests in windowing systems because of the advantages of enabling concurrent applications. We present the NEMOSHELL windowing system for multi-user interactive surfaces. We developed the system based on the Wayland system, a replacement for the Linux X. NEMOSHELL is designed to support multiple simultaneous applications, legacy input devices, legacy applications, and dynamic user interfaces. Finally, our demonstrations illustrate the potential of our system design.

Performance Analyses of Duplicated I/O Stack in Virtualization Environment

Recently, many studies have focused on eliminating duplicated I/O stack of virtualization because I/O requests of a guest machine have to pass through the I/O stack of both the guest and host machine. In this paper, we analyze the effect of duplicated I/O stack (e.g., page cache, file system, I/O scheduler, and device driver) between the guest and host machine. Especially, we study an I/O scheduler of the guest machine and the virtio of QEMU to understand how the I/O scheduler affects the overall I/O performance. Our experimental results show that Noop scheduler can lead to better performance than other schedulers, such as CFQ and Deadline. As expected, virtio shows the best performance in sequential write pattern since it eliminates the overhead caused by I/O scheduler of the guest machine by bypassing its I/O scheduler. However, in this paper, we report that the virtio reveals performance limitation in random write pattern. Our experimental results clearly show that the duplicated I/O stack significantly impacts on the overall performance of the guest machine. Especially, using virtio to eliminate the I/O scheduler of the guest also has negative impacts on I/O performance according to the write pattern.

Software-based single-node multi-GPU systems for multiple display environments

In this paper, we suggest a novel multi-GPU scheme that can provide flexible interconnections between multi-GPUs and displays with no additional hardware by implementing new software features, while avoiding system overhead. Our proposed scheme offers the complete concept of windowing system and GPU driver for single-node multi-GPU system on multiple display environments.

Page allocation scheme for anti-fragmentation on smart devices

In embedded smart devices, efficient memory management is critical because they have relatively small main memory to reduce cost and power consumption. Though Input Output Memory Management Unit (IOMMU), which is recently adopted on embedded smart devices, gives more free spaces in system memory, it increases memory allocation time when memory space is highly fragmented. In this paper, we propose a new page allocation scheme, called A-GPBM, to reduce fragmentation of anonymous pages and secure more physically contiguous pages. Experimental results show that our algorithm decreases unusable free space index for order 4 from 0.93 to 0.32.

Grouping Applications Using Geometrical Information of Applications on Tabletop Systems

In this paper, we propose a grouping scheme that classifies applications into groups for individual users by utilizing their geometrical information on a tabletop system. The proposed scheme investigates the geometrical information of each application, such as its position on the display and its rotational information, and then groups the applications of each individual user by utilizing a classifier with the geometrical information. We evaluate the proposed scheme with lab experiments, and the results show that, on average, 95.6% of applications are well classified into their users.

Software-based single-node multi-GPU systems for interactive display wall

A display wall, one of the multiple display environments (MDEs), has been widely adopted in various fields including consumer electronics. Conventional display walls are constructed by utilizing cluster-based systems where multiple displays are connected via multiple cluster nodes. However, cluster-based systems have suffered from communication overheads among the cluster nodes when they provide interaction features and run diverse applications. This paper proposes a software-based single-node multi-GPU system for interactive display walls. The proposed system allows multiple GPUs in a single-node system to connect multiple displays, by providing an individual compositor for each GPU in the system. This paper also analyzes memory management models of multi-GPU systems and compositing overheads for multiple displays. Experimental results show that the FPS (frames per second) of the proposed system outperforms that of the conventional system by up to 1.36x, reducing copy overheads among the GPUs by efficient memory management and distributing compositing overheads to multiple GPUs.

User Isolation in Multi-user Multi-touch Devices Using OS-Level Virtualization

Providing multi-user isolation is one of the most important issues in computing environments with multi-touch displays because multi-touch devices such as tabletops allow multiple users to interact simultaneously. However, existing window manager, which is commonly used to control the placement and appearance of windows, never provides isolation among the users. In this paper, we present a method that provides isolation in multi-touch multi-user computing systems using OS-level virtualization and show the effectiveness of the method with serveral experimental results. Our experimental results show that OS-level virtualization sucessfully provides both multi-user interaction and isolation in multi-touch devices.

Remote-Launch: Borrowing Secure TCB for Constructing Trustworthy Computing Platform

Commodity operating systems have become extremely large to provide a lot of services, and then their vulnerability has been targeted by malicious attack. In order to increase security in operating system, there have been many attempts to reduce the size of Trust Computing Base (TCB). However, most of the approaches have applicability limitations due to hypervisor vulnerability and additional hardware requirements. To address these limitations, instead of reducing TCB size, we propose a novel approach to enhance the security of the system. We hide secure TCB for sensitive applications, and thus build an isolated secure environment using a well-equipped infrastructure. For evaluation, we implement a prototype, called Remote-Launch, which runs a security-sensitive process on the borrowed secure TCB.

Address Space Maintaining Scheme for Fast Program Execution in Linux-based Systems

The environment of Internet of Things (IoT) wherein various devices are connected through the Internet with value-added network functions, is currently a subject of active study. Accordingly, the existing computing environment based on desktop or mobile systems is being expanded into a computing environment of more diverse devices. Because the response of program launching is important in terms of User Experience (UX) in IoT environments, the technology for guaranteeing rapid response of program launching in IoT devices is getting the focus of much current research. In this paper we analyze the Zygote technique, which is being used for faster program execution in Android systems, and, based on our results, we propose an address space maintaining scheme for the rapid launching of programs for use in Linux-based systems. Our scheme utilizes the Copy on Write (CoW) technique in Linux systems as well as the Zygote technique of Android systems. In order to evaluate the proposed scheme, we implemented our scheme on Linux systems and performed several experiments. The experimental results show that the proposed scheme shortens the launching time up to 99%, compared to the existing technique.