中央研究院 資訊科學研究所

Despite years of efforts of research communities and industry on indoor position/location technologies and services, there is still no clear winner and no common standard today. By and large, existing indoor positioning services/systems (IPS) are not well suited for large public buildings/facilities such as train stations, airports, major hospitals, department stores, sports centers and theater complexes. Common characteristics of such buildings include complex and dynamic operating environment, large/fast fluctuations in density of people, criticality of IPS during emergencies and diversity in capabilities of user devices. Existing IPS do not address these challenges well enough to be easy to deploy and maintain, stay scalable in response to orders of magnitude surge in location queries, degrade gracefully to be disaster resilient, and minimize requirements of user devices to use the service. This report describes an IPS, called BeDIPS (Building/environment Data-based Indoor Positioning Service). It is designed to meet these high-level requirements of IPS for large public buildings. As its name implies, the service relies on a building and environment data/information cloud (BeDIC), which among other types of data, contains 3D coordinates and geometric models of every object of interest. In particular, it contains the coordinates of location beacons (called LBeacons), which the service uses to deliver location information to users. Basically, Lbeacons are a low-cost, Bluetooth Smart Ready device. At deployment and maintenance times, the 3D coordinates of every Lbeacon are retrieved from BeDIC and stored locally. Once initialized, each Lbeacon broadcast its coordinates to HereUAre, a simple messaging application on Bluetooth enabled mobile device nearby. Interferences among Lbeacons are minimized by partitioning the network of Lbeacons into subnets and having beacons in each subset transit in a time division multiplex manner. From the coordinates of all Lbeacons heard by the HereUAre on a device, the application can easily, and sufficiently accurately, estimate the coordinates of the device when the network of Lbeacons are sufficiently dense and well placed. The report presents the architecture and design of BeDIPS and discusses reasons for its feasibility.

Full system emulators provide virtual platforms for several important applications, such as kernel and system software development, co-verification with cycle accurate CPU simulators, or application development for hardware still in development. Full system emulators usually use dynamic binary translation to obtain reasonable performance. This paper focuses on optimizing the performance of full system emulators. First, we optimize performance by enabling classic control transfer optimizations of dynamic binary translation in full system emulation, such as indirect branch target caching and block chaining. Second, we improve the performance of memory virtualization of cross-ISA virtual machines by improving the efficiency of the software translation lookaside buffer (software TLB). We implement our optimizations on QEMU, an industrial-strength full system emulator, along with the Android emulator. Experimental results show that our optimizations achieve an average speedup of 1.92X for ARM-to-X86-64 QEMU running SPEC CINT2006 benchmarks with train inputs. We use a set of real applications downloaded from Google Play as benchmarks for the Android emulator. Experimental results show that our optimizations achieve an average speedup of 1.42X for the Android emulator running these applications.

Recently, asymmetric multi-core architecture have become an important issue in CPU design, software scheduling, and virtualization. In a virtualization environment, a hypervisor scheduler assigns virtual cores to physical cores for task execution. However, a load-balancing scheduling strategy for a symmetric multi-core platform (SMP) is unaware of core asymmetry. The deployment of such a strategy to an asymmetric platform may cause performance degradation or energy waste. To take full advantage of the improved power efficiency and performance made possible by asymmetric multi-core platforms, we need a new scheduling strategy. In this paper, we propose an energy-efficient, asymmetry-aware scheduling mechanism for hypervisors on asymmetric multi-core platforms. The goal is to generate an energy-efficient scheduling plan with guaranteed performance.

Agent-Based Disaster Simulation Environment (ABDiSE) is a framework that provides model elements and tools to support modeling and simulation of common types of natural disasters, including fires, floods and debris flows. The underlying disaster model is agent based: Active objects describe how agents move, attach, and interact with each other and with their environment. ABDiSE is extensible: New agent types and external simulators needed to model elements and dynamics of new disaster scenarios and define behaviors and interactions of agents can be added without requiring revision and recompilation of the framework. ABDiSE is a multi-threaded, capable of taking advantage of available computing resource to speed up simulation.