Our research on systems and networking is guided by the vision that these two areas influence each other strongly. Faculty members constituting this group have wide spanning research interests that cover the conceptual as well as experimental aspects of computer systems and networking. The group attracts a substantial amount of research funding and has a strong presence internationally through publications, invited tutorials and memberships in the technical programme committees of leading conferences. The major themes and research areas are:

Systems

The major areas currently being investigated by the Systems group are:

Grid Computing

In recent years, we have seen an increasing number of Internet-scale applications that focus on resource sharing. The technologies that these applications leverage include peer-to-peer, grid and web services. In this stream of work, we focus on various aspects of resource sharing. In computational resource sharing, we have developed a Javabased grid middleware called ALiCE (Adaptive and scalable internet-based Computing Engine). On infrastructure for resource sharing, we are making notable contributions in three main areas: (1) a read-only DHT (Distributed Hash Table)-based resource discovery and indexing technique, (2) a game theoretic market-based resource pricing and allocation technique, and (3) a service provisioning method for large distributed simulation on the grid. In the area of software component sharing, we are developing a framework consisting of semantic composability theory and validation techniques for large-scale sharing of simulation model and components.

Embedded Computing Systems

Embedded computing systems are increasingly playing a dominant role. In particular, they are beginning to deliver more powerful services in application domains such as aviation, military, telecommunications and process control. Due to the falling price of hardware and the trend towards ubiquitous networking, embedded computing is also becoming a key component of many common place applications and household products.

In these domains, lay consumers typically demand a level of reliability and predictability not normally expected from traditional computer software. This calls for a design process that is effective and reliable but yet has a short turn-around time. Such a design process should encompass multiple levels of systemdescriptions and the means for going from one levelof description to a neighbouring one while preserving behaviour and integrity of design.

Our research activities in this area include: model driven design methods with a focus on UML-based notations and SystemC-based simulations, timing and schedulability analysis of embedded software, performance and power analysis of real-time audio, video and 3D game applications, performance modelling of automotive software and protocols and, finally, tools and techniques for reconfigurable computing. Our research is presented at the leading Electronic Design Automation (EDA) conferences around the world.

The faculty members involved in systems research are:

• CHAKRABORTY Samarjit
• MITRA Tulika
• PUNG Hung Keng
• ROYCHOUDHURY Abhik
• TAN Soon Huat,Gary
• TEO Yong Meng
• THIAGARAJAN P.S.
• WONG Weng Fai

Networking

The Communication and Internet Research Lab (CIRL) (http://www.cir.nus.edu.sg/) and the Network Systems and Services Lab (NSS) (http://lucan.ddns.comp.nus.edu.sg/) constitute the hub of our research activities in networking. The group is presently involved in the following research activities:

Quality of Service Management

To address Quality of Service (QoS) issues arising from the heterogeneity of end-hosts and networks, we envisage a holistic approach to QoS control and management in grids which integrates QoS mechanisms in the applications, middleware and networks. In our approach, the QoS of network, host’s middleware, grid middleware and application’s are modelled as configurable QoS components (meta models). The way the meta-data are represented and stored is QoS Meta-Information Based (QMIB). This encompass two major categories of information, namely the application QoS information and the grid service QoS information. The potential contributions of our work are a more expressive extensible semantic QoS Specification schema (SQS) for describing QoS requirements of applications and services, an intelligent algorithm for the mapping of QoS concepts and requirements to QoS specifications at the resource level (stored as QMIB), and the QoS runtime for supporting end-toend QoS over existing grid middleware and network infrastructures.

TCP Performance

In this area of research, we study the dynamics of the TCP protocol and its adaptation to the emerging wireless networks which are known for high loss and long round-trip time. We have just initiated a project called “Reactive Transport Service” in which we intend to do basic research to understand the dynamics of transport layer protocols in the context of the mobile wireless environment of an All-IP converged network and the characteristics of access networks and mobility pattern of Internet enabled devices. We aim to develop heuristics that could be applied to the design of algorithms capable of learning and reacting to the heterogeneous and dynamic nature of the network.

Sensor Networks

With the capability to construct low power and low cost devices that support sensing, computation and communication capabilities, it is possible to create new and exciting types of applications in a wide range of areas including transportation, manufacturing, health care, environmental oversight, and safety and security. Due to the constraints of sensor nodes, many networking issues such as MAC, network and transport protocols have to be revisited. In particular, protocols for sensor networks have to be very energy efficient. One way to reduce energy consumption is to apply the idea of cross-layer design. Here, the physical, MAC, network and transport protocols are designed in an integrated way so that better energy efficiency can be obtained. We are also interested in building a novel and practical sensor network system to guide and verify the design.

Heterogeneous Wireless Networks

Development in new radio technologies and increase in user demand are driving the deployment of a wide array of wireless networks, ranging from 802.11 networks in the local area, to third generation data-only wireless networks in the wide area. Such a diverse set of wireless networks poses a number of interesting issues. We are mainly interested in resource management related to heterogeneous wireless networks. We consider the case where multiple wireless networks, including 802.11, 802.16 and 3G data-only networks are being deployed, and a single wireless provider owns and operates these wireless networks. We are interested in how a network manager can efficiently manage the common pool of radio resources, and how the quality of service of real-time applications (such as voice) and completion time of short data transfer can be improved through redundant transmission.

Data Availability of Higher-level User Tasks

Data replication is a key technique for ensuring data availability in the presence of machine failures. Most previous work on data availability focuses on the availability of a single data object (e.g., a single file). A higher-level user operation, however, often needs to access multiple data objects to perform a certain task.Our research reveals that the relative placement or assignment of replicas of different objects can result in subtle correlation among objects and dramatic effects on system availability.

Using strong probabilistic tools such as Janson’s inequality from random graph theory, we are able to find the optimal assignment under different settings.We have also proposed practical decentralised designs to approximate these optimal assignments in practice, and implement the design in a prototype wide-area storage system called MOAT. PlanetLab deployment and simulation results show that using the appropriate assignment often decreases the failure probability of operations by multiple orders of magnitude.

Defending against Sybil Attacks in Decentralised Distributed Systems

Peer-to-peer and other decentralised distributed systems are known to be particularly vulnerable to sybil attacks. In a sybil attack, a malicious user obtains multiple fake identities and pretends to be multiple distinct nodes in the system. By controlling a large fraction of the nodes in the system, the malicious user is able to “out vote” honest users in collaborative tasks such as Byzantine failure defences. We have proposed SybilGuard, a novel protocol for limiting the corruptive influences of sybil attacks. SybilGuard uses as its basis the social network among user identities, where an edge between two identities indicates a human established trust relationship. Malicious users can create many identities but few trust relationships. Thus, there is a disproportionately-small cut in the graph between the sybil nodes and the honest nodes. SybilGuard exploits this property to bound the number of identities a malicious user can create.

The faculty members involved in networking research are:

• ANANDA Akkihebbal L.
• CHAN Mun Choon
• LEONG Wing Lup, Ben
• PUNG Hung Keng
• YU Haifeng