Soo-Cheol Oh

An Efficient Linux Kernel Module supporting TCP/IP Offload Engine on Grid

All resources of grid systems are connected by network and Ethernet-based TCP/IP is a widely used network in the grid systems. The TCP/IP is processed on a host CPU in computer systems and this imposes heavy loads on the host CPU. Recently TCP/IP offload engine (TOE) technology, which processes TCP/IP on a network adapter instead of the host CPU, has become an important approach to this problem. To adopt the TOE to computer systems successfully, it is necessary to develop a kernel module supporting the TOE. The TOE kernel module should provide the binary compatibility for the BSD socket interface, which is widely used in the most network programming, and support both of the TOE and the Ethernet-based NIC simultaneously. This paper proposes a TOE kernel module satisfying these requirements in the Linux systems. Experimental results also show that the TOE kernel module imposes little load on the host CPU

Fast Operating System Switcher for mobile CE devices

This paper proposes a fast Operating System (OS) Switcher for ARM-based Consumer Electronics (CE) such as Smart Phones, Tablets and IPTVs. The OS switcher enables users to utilize multiple OSes on single H/W device without security problem from interference among OSes. The OS switcher is notably faster than the multi-boot approach and can preserve the previous working states. The OS switcher does not burden any run time overhead compared to virtualization and can be easily applied to the new product. We implemented the OS switcher in the bootloader layer without any modification of OS.

A Network I/O Architecture for Terminal-Initiated Traffics in an Ubiquitous Service Server

In ubiquitous environment, terminal-initiated traffics(e.g. , sensor data streams) will become more popularized than human-initiated traffics(e.g. , web browsing, file transfer, and media streaming). The characteristics of the terminal-initiated traffics can be described as the streams of small sized packets with the large number of simultaneous connections. Thus, an ubiquitous service server will suffer from the burden of the terminal-initiated traffics since the per-packet-cost of a CPU will increase proportionally with the number of terminals. In this paper, we propose Latona architecture that offloads not only TCP/IP processing but also parts of socket processing from a host to minimize the per-packet-cost of a CPU in a Linux server. Based on the experimental results, the Latona kernel could save 50% and 79% of the kernel execution time of send() and recv() for 32 bytes transfer in the legacy TCP/IP stack. The packet-per-second were 21.9K and 18.3K for send() and recv() ,respectively. The bandwidth increased as the size of payload increased. The profile of detail execution time showed that the bottleneck of the Latona was for handling the socket and the TCP in Latona hardware.

Reducing CPU Power Cost of Cluster System

Cluster systems with high price-to-performance ratio have been utilized in various fields and size of cluster systems is increasing rapidly. Also, as the information retrieval and sharing through the Internet are used widely, scale of portal sites that collect, process and provide information, become larger. These portal sites have large scale cluster system to serve mass information, and operation and maintenance of the large cluster system is a big challenge to the portal sites. Because power cost is one of the main issues in the maintenance of the large cluster system, it is necessary to reduce the power cost of the cluster system. As one solution for this problem, this paper presents the CPU power management mechanism that reduces power usage of the cluster system.

Fast Provisioning Service in Shared-Storage Based VDI According to User’s Type

The main purpose of VDI (Virtual Desktop Infrastructure) provisioning is to provide the virtual desktop and the resources that the customer wants to use quickly. The customer of DaaS appears in various forms according to the type of user. The most commonly categorized approach can be broken down depending on whether the user permanently owns the virtual desktop or whether the virtual desktop is allowed for temporary use. And the type of user will be different according to the provided policy. In this paper, we provide a mechanism to allocate virtual desktops to users in 320 ms period of time according to the form of various users, and provide a fast provisioning method by applying them to a shared-storage based DaaS system.

ARM-based thin virtual desktop client integrating electrical power control with monitor

This paper proposes an ARM-based thin virtual desktop client integrating electrical power control with monitor. User can turn on/off both of client and monitor using one monitor power button. This client consists of client power controller and ARM-based virtual desktop client. The client power controller performs integrated power control. The controller detects power-on and power-off events of the monitor and controls power of the client. For power-on event, the client power controller turns on directly the power line of the client. For power-off event, the controller controls both of the power line and power management SW module of the client. The size of the client is compact and mountable to backside of the monitor with VESA mount.

LATONA: an advanced server architecture for ubiquitous sensor network

The emerging Ubiquitous Sensor Network (USN) makes connection less datagrams and short event packets get popular. A large number of short term event packets of USN can cause serious problems, such as interrupt handling overhead and context switching overhead. Furthermore, heavy load of the packet security methods needs enough processing power. Then, the more USN develops, the more network overheads would be loaded into host CPU. To solve the problems, we propose a special server component including TOE (TCP/IP Offloading Engine) and H/W IPSec (IP Layer Security) for USN.