Institute of Information Science Academia Sinica
Institute of Information Science Academia Sinica
Network System and Service Laboratory
Principal Investigators:
:::Ling-Jyh Chen (Chair) :::Meng-Chang Chen :::Sheng-Wei Chen
:::Wen-Tsuen Chen :::Tyng-Ruey Chuang :::Jan-Ming Ho :::Jane W. S. Liu

Our research addresses several aspects of network systems and services, including designing participatory sensing systems for environmental monitoring, developing critically needed information and communication technologies for disaster management, improving user experience for audio streaming, supporting memory-based computation for large-scale de novo genome assembly, and developing the key components of next-generation cellular networks, such as Software-Defined Network (SDN), Network Function Virtualization (NFV), Internet of Things (IoT), and Massive MIMO.

1. Ling-Jyh Chen
We study the Internet of Things (IoT) and participatory sensing systems, and have combined the two concepts to build a large scale system, called AirBox, for fine particulate matter (PM2.5) monitoring. The project engages citizens to participate in environmental sensing and enables them to make low-cost PM2.5 sensing devices on their own. It also facilitates PM2.5 monitoring at a finer spatio-temporal granularity and enriches environmental data analysis by making all measurement data freely available. As of mid-2017, we have deployed more than 2,000 devices in over 30 countries, and we have established collaborations with international and domestic researchers on several key issues regarding IoT systems and the interdisciplinary topic of environmental monitoring systems.
Specifically, we have investigated low power wireless area networks (LPWAN) techniques in the AirBox design, and we have studied IoT security issues to ensure confidentiality and integrity. Moreover, we have been working closely with environmental researchers and authorities on sensor calibration under different application scenarios. Using the AirBox measurement data, we have developed a set of algorithms to rank the confidence levels of devices, detect ongoing PM2.5 emissions, and assess device attributes (e.g., deployed indoors or nearly-consistent emission sources). Our ongoing work is to improve spatio-temporal data analysis and identify additional intrinsic properties of PM2.5 distributions from real time measurement data, and to conduct short-term PM2.5 concentration forecasting at a finer spatio-temporal granularity with a good accuracy.

2. Jane W. S. Liu
We continue to collaborate with several research fellows and faculty members in earth science, urban planning and computer science on information and communication technologies for disaster management. At the start, our focus was on a framework for building open and sustainable disaster management information systems. Its elements include a responsive, trustworthy emergency-information brokerage service; disaster resilient heterogeneous, plug-n-play networks and dynamic logical information exchange; models and tools for fusing data from people and intelligent things; and building blocks and infrastructure for active use of disaster alerts. Prototypes of these elements have demonstrated their concepts and feasibility. For example, an information delivery middleware over interwoven heterogeneous networks was prototyped to demonstrate the concept of open information gateway.
Our current efforts emphasize the generation and use of data. One thrust is directed toward developing methods and tools for generation and collection of data needed for disaster risk reduction. Examples include CROSS (CROwdsouring Support system for disaster Surveillance). A typical disaster surveillance system must assist in making critically important decisions within minutes or hours before disasters strike. When it is necessary to crowdsource human sensor data, the emergency manager needs help in selection of participants, plan tours for them to cover locations where observational data are needed, and fuse data from them in real-time with physical sensor data to improve the quality of coverage. CROSS was built to meet these needs. We have solved realistic variants of underlying participant selection, tour planning and symbiotic data fusion problems and produced solutions not only of practical utility to CROSS, but also of theoretical significance. We plan to integrate these components into the well-known platform Ushahidi. CROSS is currently being used by our earth science colleagues to coordinate trained volunteers after each significant earthquake in Taiwan to collect data on new geo-hazards in order to assess risks of earthquake-triggered compound disasters. Another major thrust of our work is at developing active smart embedded devices, mobile applications and services/systems in smart homes and buildings, which can automatically process alerts from authorized senders and building safety systems and take location specific actions to minimize chance of injury and reduce property damages when disasters strike. Each device (or application) selects its action (or response instructions) in response to an alert based on the type and parameters of the alert and attributes of the building, interior layout and nearby objects around the device (or application). Our work to date has demonstrated that easy to customize and maintain active smart devices/applications for diverse purposes can be built on a common architectural framework from reusable components. Moreover, alert messages that are pushed asynchronously over the Internet can meet the end-to-end delay requirements of time-critical alerts. Our experimentation with a prototype active emergency response system in an office building during a simulated strong earthquake has demonstrated the effectiveness of such systems.
In order to enable the pervasive use of active devices and applications, we are developing a building/environment data and Information (BeDI) system (or BeDIS) as a part of the infrastructure needed to support location-specific, active emergency preparedness and response within large public buildings. One component of BeDIS is an indoor positioning system (IPS) that is unique among existing IPSs. Our IPS can reliably deliver location data with 3-5 m or 5-10 m horizontal accuracy to both smart phones and most legacy Bluetooth devices. It is scalable during orders of magnitude surges in crowd density and does not require Internet to function. Further, it degrades gracefully when parts of it are damaged and is easy to deploy and maintain. The other major component of BeDIS is BeDI mist, which is a virtual repository of data on the building, interior layouts and facilities. It is capable of delivering fine-scale, location-specific decision-support data and emergency response instructions to hundreds and thousands of active devices and people. Prototypes of these components are ready for experimental use. We are planning pilot studies to assess usability and effectiveness of BeDIS and active applications supported by it in representative large public buildings.

3. Jan-Ming Ho
We are interested in the following problems:
1. Improving user experience with DASH
Dynamic Adaptive Streaming over HTTP (DASH), also known as MPEGDASH, is a popular video streaming protocol over HTTP. DASH enables high quality video streaming using conventional HTTP web servers without the need of an additional streaming server. A DASH video is divided into segments of a constant duration, usually a couple of seconds. Each video segment is then encoded into several small files that may be played back at a specific bit rate. The client can then use current network conditions to adaptively select appropriate segment size in real time. It was drafted by MPEG as an international standard in 2011 and published by ISO/IEC as ISO/ IEC DIS 23009-1.2 in 2012.
DASH is designed to dynamically react to network bandwidth and has been shown to perform well in mobile environments for which the available bandwidth is difficult to predict. However, if the available bandwidth changes are dramatic (for example, passengers on an underground subway or a THSR train may receive large bandwidth at a train station, and total cessation in a tunnel), video streaming may be problematic. A trivial solution to improve the quality of video streaming is to install sufficient number of wireless base stations along the railway. On the other hand, if the available bandwidth can be predicted, a smart scheduling algorithm might also improve the user experience.
In this study, we aim to improve quality of experience for DASH video streaming. We will crowdsource time series data of the mobile network bandwidth along several routes of Taipei Metro, also known as Taipei Rapid Transit System, and develop models. We will further design efficient prediction and scheduling algorithms to optimize video streaming experience for subway riders. The crowdsourced data will be used to validate efficacy of the models and algorithms.

2. Architecture design to support memory-based computation of large-scale de novo genome assembly using MapReduce framework
In processing NGS reads of large genomes for de novo genome assembly, using the MapReduce computing framework, we must generate suffices for each read in order to perform sorting. The number of tuples, generated to hold suffices, ranges from 100 to 400 and equals the length of the read. In other words, for a 100GB NGS data file, the total size of tuple data is 10TB. To process this amount of data in memory, we may either allocate a huge number of computing nodes (on the order of 1000) or we may separate storage nodes from the computing nodes. In the latter case, we may reduce the total number of nodes, both for storage and computing, by an order of magnitude. Notably, in some environments, the number of available computing nodes may be limited. In such cases, it is I/O bound that each computing node must process a data set much larger than the size of its physical memory, which is time consuming.

In this study, we aim to enhance the MapReduce framework to support applications that require computation of suffix arrays. We thus have two sub-goals. One is to implement a suffix-array algorithm, based on the notion for separation of storage and computing, to reduce the total number of processors. We will also design and develop memory-based storage architecture to support applications to be implemented with separate storage and computation.

4. Wen-Tsuen Chen
The 5th generation mobile communications system (5G) has attracted much attention recently. As such, many task groups have been established to work toward the identification of system requirements, novel system concepts and potential access technologies for 2020 and beyond. In order to produce a functional 5G network, we must overcome major challenges with regard to explosive growth of mobile data traffic volume, number of connected devices, and typical end user data rate.
In future 5G networks, the deployment of ultra-dense smallcell networks is a feasible and reasonable solution for enhancing spectrum utilization to meet 5G requirements. Toward management of 5G networks, softwaredefined networking (SDN) decouples the software defined control plane from the hardware driven data plane on general purpose hardware. Because there is an increasing trend towards implementing more functions of mobile communications systems in software, SDN will be an essential strategy for 5G management.
Many challenges and problems in 5G networks still need to be resolved. We focus on mobility and resource management mechanisms in SDN-based ultra-dense smallcell networks to seek possible solutions. In the mobility management aspect, we study the intra-domain and inter-domain handover procedures in an SDN-based ultradense smallcell network. Moreover, we consider both the handover decision and group mobility issues. In the resource management aspect, we explore advanced techniques for improving the network throughput, such as massive MIMO, CR, CoMP, etc. Finally, we also take the fairness and QoS guarantee into account..

5. Wen-Tsuen Chen
We perform research in the field of next-generation cellular networks, known as 5G networks, which employ technologies such as Software-Defined Network (SDN), Network Function Virtualization (NFV), Internet of Things (IoT), and Massive MIMO. Our research objectives include traffic engineering, resource optimization, and security in a promising network architecture.
Current traffic engineering in SDN is mostly accomplished with unicast methods. By contrast, multicast can effectively reduce network resources consumption to jointly serve multiple clients. We investigate a reliable multicast tree for SDN to minimize both multicast and recovery costs for reliable multicast transmission. We also study multicast traffic engineering for multiple trees, which is very challenging because we must jointly consider the bandwidth consumption minimization of a single multicast tree, the link capacity for flows, and the node capacity for storing the forwarding entries in Group Table. In addition traffic engineering, there are inherent benefits to incorporating SDN and NFV in the next-generation cellular network, particularly for the management and orchestration of VNFs. We design a scheme of service chain embedding to maximize the total amount of flows, while bounding the process overhead of the flows on a node, by its computation capability. The total amount of flows is on a link by its bandwidth capacity. Furthermore, most previous studies on NFV focus on unicast service chains and thereby are not scalable to support a large number of destinations in multicast. We make the first attempt to tackle the new, and challenging problem of constructing a service tree that contains multiple branching service chains required by each destination.
Moreover, we also study several security issues in SDN/NDF architecture. Deep packet filtering (DPF) has been demonstrated to be an essential technique for effective fine-grained access controls, but it is commonly recognized that the technique may invade the individual privacy of the users. Fortunately, secure computation can address the tradeoff between privacy and DPF functionality. However, the current solutions limit the scalability of the network due to the intensive computation overheads and large connection setup delay. We propose a privacypreserving deep packet filtering protocol that can efficiently perform filtering functions on encrypted traffic while decreasing the communication overhead and setup delay for the controller in SDN. In order to mitigate HTTP DDoS attacks, shuffling-based moving target defense redirects user traffic among a group of virtualized service functions and has been regarded as one of the most effective strategies. Nevertheless, previous work did not note that frequent changes of user traffic will significantly intensify the control overhead of SDN. Therefore, we have developed a cost-effective shuffling-based defense system with a guaranteed performance bound.
The number of Internet of Things (IoT) devices is rapidly increasing and has become quite extensive. Thus, the IEEE 802.11ah standard adopts the grouping-based MAC protocol to reduce the contention overhead of each group of devices. However, most existing designs randomly assign devices to groups, and little attention has been paid to the problem of forming efficient groups. We propose a load-balanced grouping algorithm to improve channel utilization for each group. In addition, Multi-User Multiple Input Multiple Output (MUMIMO) enables a multi-antenna access point (AP) to serve multiple users simultaneously, and has been adopted as the IEEE 802.11ac standard. In practice, user frames with heterogeneous lengths may cause the concurrent transmission opportunities to become incompletely utilized. To resolve this inefficiency, we present a PHY-MAC design that adds additional frames to fill up the idle channel time and better utilize the spatial multiplexing gain. On the other hand, IoT services may be associated with burst traffic, critical tasks, and low latency requirements. To this end, we propose to utilize idle devices in IoT networks to boost the transmission data rate for critical tasks through multiple concurrent transmissions. Furthermore, we also study the specific applications of IoT. For example, wireless surveillance in cellular networks has become increasingly important. We design an efficient method to minimize the number of allocated resource blocks (RBs) while guaranteeing the coverage requirement for surveillance systems in uplink networks.