Institute of High Performance Computing

With advancements in computational resources, scientists, business analysts, healthcare system analysts and economic forecasters are being provided ample opportunities of solving their multiple cross disciplinary problems with higher level of detail and accuracy, and within reasonable timeframe. This has led to a massive increase in the amount of data being generated. Large-scale data come in many forms from almost all practical domains – examples include Terabytes-per-minute production-line data from industrial engineering and millions of cell images from bioinformatics. Knowledge discovery from these large repositories poses domain-specific challenges that require considerations beyond generic data mining-based strategies. At ASA, we approach large-scale data mining from a practical perspective that is knowledge-centric. First we collaborate closely with domain experts to resolve problems using solutions that are domain-related, second abstract elements from the solutions that are reusable, and third compile the abstracted modules into an integrated knowledge-discovery service-platform and construct a library of large-scale data mining capabilities. We are currently building data mining solutions including for visual-summarisation of cell phenotype patterns, supercomputing centre log analysis, and fault-detection of devices at a manufacturing site. Complementary to these activities, we conduct research on algorithms for large-scale domain adaptation to support transferable data mining models and will incorporate HPC techniques to generate an infrastructure for housing parallel data mining algorithms that utilise multicore and GPU architectures. Our research puts strong emphasis on scalability and meaningful knowledge representation.

The types of multi-disciplinary problem that we have worked on include weather prediction, text mining, biomedical data analysis, workload modelling, resource mapping for various HPC architectures, and infectious disease modelling. Taking infectious disease modelling as an example, it is well-established that emerging and re-emerging infectious diseases can cause global pandemics. Faced with such grave challenges, we are investigating epidemiological modelling, simulation and analysis of infectious diseases, and have studied dengue, avian flu, hand, foot and mouth disease, etc. The infectious disease research group has developed contact-network-based epidemiological models for investigating infectious disease spread patterns that explore various techniques of data analysis in graph networks, modelling and simulation based on data mining and high performance computing techniques. Through this effort, we attempt to determine the feasibility and effectiveness of implementing various control strategies. These research outcomes may eventually help assist public health decision makers in shaping their immediate or future policies.

Our collaborators for these research projects include University of British Columbia of Canada, National University of Singapore, Nanyang Technological University, Ministry of Health, National Environment Agency, Texas Advanced Computing Centre, Sun Microsystems, and several other A*STAR research institutes.

Awards

• Wu Yan, Gary Lee, Fu Xiuju and Terence Hung, “Detect climatic factors contributing to dengue outbreak based on wavelet, support vector machines and genetic algorithm”, International Conference on Data Mining and Knowledge Engineering of World Congress on Engineering, 2008. Awarded the Certificate of Merit and invited as a book chapter by Springer.

Recent Publications:

• Mark J. Schreiber, Edward C. Holmes, Swee Hoe Ong, Harold Soh, Wei Liu, Lukas Tanner, Pauline P.K. Aw, Hwee Cheng Tan, Lee Ching Ng, Yee Sin Leo, Jenny G H. Low, Adrian Ong, Eng Eong Ooi, Subhash G, “Genomic Epidemiology of a Dengue Virus Epidemic in Urban Singapore”, Journal of Virology Vol 83(9): 41634173 May 2009
• Harold Soh, Yew Soon, Quoc Chinh, Quang Huy Nguyen, Mohamed Habibullah; Terence Hung and Jer-Lai Kuo, “Discovering Unique, Low-Energy Pure Water Isomers: Memetic Optimization and Landscape Analysis”, IEEE Transactions on Evolutionary Computation (accepted).
• Thorsten Matthias Riechers, Shyh-hao Kuo, Rick Siow Mong Goh, Harold Soh, Terence Hung, Abid M. Malik, "A Case Study on Dynamic Kernel Adaptation in a Component-based Infectious Disease Simulator", Workshop on Component-Based High Performance Computing (CBHPC 2009) at Supercomputing Conference (SC) 2009.
• Qin Zheng, B. Veeravalli, C. K. Tham, ”On the Design of Fault-tolerant Scheduling Strategies Using Primary-Backup Approach for Computational Grids with Low Replication Costs”, IEEE Transactions on Computers vol. 58, no. 3, Mar. 2009.
• Yi Wen Wong, Rick Siow Mong Goh, Shyh-hao Kuo, Malcolm Yoke Hean Low, “A Tabu Search for the Heterogeneous DAG Scheduling Problem”, International Conference on Parallel and Distributed Systems 2009.
• Tianyou Zhang, Xiuju Fu, Rick Siow Mong Goh, Chee Keong Kwoh, and Gary Lee, “A GA-SVM feature selection model based on high performance computing techniques”, IEEE International Conference on Systems, Man, and Cybernetics 2009.
• Tuan Zea Tan, Rick Siow Mong Goh, Verdi March, and Simon See, “Data Mining Analysis to Validate Performance Tuning Practices for HPL”, IEEE International Conference on Cluster Computing (Cluster) 2009.
• Tuan Zea Tan, Rick Siow Mong Goh, Verdi March, and Simon See, “A case study of speeding up performance tuning process using data mining approach”, International Conference on Scientific Computing 2009.
• Xiuju Fu, Sonja Lim, Lipo Wang, Gary Lee, Stefan Ma, Limsoon Wong, Gaoxi Xiao, “Key Node Selection for Containing Infectious Disease Spread Using Particle Swarm Optimization”, IEEE Swarm Intelligence Symposium 2009.
• Tuan Zea Tan, Verdi March, Rick Siow Mong Goh, and Simon See, "Performance tuning of parallel application using computational intelligence analysis", HPC Asia 2009.
• Tianyou Zhang, Harold, Soh, Xiuju Fu, Limsoon Wong, Stefan Ma. “HPCgen - A Fast Generator of Contact Networks of Large Urban Cities for Epidemiological Studies”, International Conference on Computer Modelling and Simulation 2009.
• Quoc Chinh Nguyen, Yew Soon Ong, Harold Soh and Jer-Lai Kuo, “A Multi-scale approach to explore the potential energy surface of water clusters (H2O)nn ≤ 8”, Journal of Physical Chemistry A Vol. 112, No. 28, pp. 6257-6261, June 2008.
• Xiuju Fu, Wei Jian Ian Teo, Harold Soh, Gaoxi Xiao, Gary Lee, Terence Hung, Limsoon wong, Stefan Ma, “Simulating Avian Influenza Spread over Singapore Clinic Networks”, 10th International Symposium on Respiratory Viral Infections 2008.
• Xiuju Fu, Christina Liew, Harold Soh, Gary Lee, Terence Hung, Lee-Ching Ng, “Time-series infectious disease data analysis using SVM and genetic algorithm”, IEEE Congress on Evolutionary Computation, 2007.

HPC-as-a-Service. Provisioning HPC resources on the fly requires more than just virtualization technologies. HPC environment can be configured in different variations. Moreover, the provisioning process needs to take care of user management, network configuration, filesystem and data integration cum management, software deployment as well as integration to user-friendly portals. As various steps are involved, automation is needed to deliver quick service availability and to reduce errors. At A*STAR, we have developed a set of tools (HPCaaST) to efficiently enable HPC on-demand service. It can be used to provide HPC environments on demand for either resources within the enterprise (private cloud) or for resources within an ISP’s data centre (public cloud).

Adaptive Infrastructure & Service Quality Management. Nowadays, supercomputer centres comprise of larger amount of compute nodes with homogeneous and/or heterogeneous mulit-core processors. The current Rank 1 system, for example, includes 129600 cores. As the size continually grows, the Mean Time to Failure (MTTF) can drop to a few hours, even minutes. This influences service performance significantly. Moreover, big spikes of demands are common for HPC services.
We have taken care to ensure that the infrastructure is designed to be flexible in order to support different integration and customization needs. Efficient scheduling, load balancing, fault tolerance strategies are designed to ensure effective service quality management. The suite of features ensures that users can have access to a comprehensive platform that can enable end-to-end high performance computing needs.

Service Discovery & Agent-based Negotiation Technologies. The paradigm of Service-Oriented Computing (SOC) signifies automation of service delivery which is of great practical value. Negotiation, as a key coordination mechanism for service providers and clients to interact and reach agreements would be a “bottleneck” in the entire automated process, if otherwise manual. We have designed and implemented software negotiation agent prototypes utilizing theory-informed human wisdom on negotiation tactics and adaptive learning algorithms towards being capable and reliable “expert agents” in negotiations. In collaboration with our industry partners (e.g., Hewlett-Packard Labs), the team is actively engaged in designing, implementing, and experimenting robust and flexible negotiation agents that will eventually be put into market use.

Operational Risks & Economic Considerations. In a competitive HPC service market of many Cloud offerings, Cloud providers need to offer suitable prices to secure their customers and maximize the projected Return on Investment (ROI). Hence, novel pricing strategies needs to be designed for a Cloud-based HPC service market. We study various HPC user characteristics (e.g., workload, user profile, service needs, and budget constraints), and important provider-specific factors and operational risks (e.g. Service Level Agreement violation, inaccurate workload prediction, insufficient capacity planning, underutilised infrastructure, and service failures).

Representative Publications (2009-2010):

• Zheng, Q., B. Veeravalli, and C. K. Tham, (2009).“On the Design of Fault-tolerant Scheduling Strategies Using Primary-Backup Approach for Computational Grids with Low Replication Costs," IEEE Transactions on Computers, Vol. 58, Issue 3, March 2009.
• Li, X., Veeravalli, B., Li, H. (2009).“Multimedia Service Provisioning and Personalization on Grid-Based Infrastructures: Now and the Future”, IEEE multimedia, Vol. 16, Issue 1, Jan.-March 2009.
• Zheng, Q. (2010). “Improving MapReduce Fault Tolerance in the Cloud”, accepted by the 15th IEEE Workshop on Dependable Parallel, Distributed and Network-Centric Systems (DPDNS), in conjunction with IEEE IPDPS, Atlanta, USA, April 2010. forthcoming.
• Zheng, Q. (2010). “Runtime Adaptation for DAGs with Uncertainties in Heterogeneous Computing Systems”, accepted by the 19th International Heterogeneity in Computing Workshop (HCW), in conjunction with IEEE IPDPS, Atlanta, USA, April 2010. forthcoming.
• Yang, Y., Singhal, S., and Xu, Y. (2009). “Offer with Choices and Accept with Delay: A Win-Win Strategy Model for Agent-Based Automated Negotiation”, 30th International Conference in Information Systems (ICIS2009), Phoenix, Arizona, USA, December 15-18, 2009.
• Yang, Y., Singhal, S., Santosa, R. Yan, Y., Foo, Y. (2009). “Negotiation as a Service: An Agent-Based Negotiation Engine for Online Business”, 19th Annual Workshop on Information Technology and Systems (WITS2009), December 14-15, 2009, Phoenix, USA. p. 224.(Best Prototype Award)
• Li, X., Singhal, S., Hung, T. (2009). “A User-centric Dynamic Cluster Partitioning Approach for HPC Service Optimization”, 28th IEEE International Performance Computing and Communications Conference (IPCCC), Phoenix, Arizona, USA, Dec 14-16, 2009.
• Yang, Y., and Singhal, S.(2009). “Designing an Intelligent Agent that Negotiates Tactfully with Human Counterparts: A Conceptual Analysis and Modeling Framework”, In Proceedings of the 42nd Hawaii International Conferences on System Sciences (HICSS2009), Track 1, Big Island, Hawaii, January 5-8, 2009.
• Yeo, C.S. and Buyya, R. (2009).“Integrated Risk Analysis for a Commercial Computing Service in Utility Computing,” Journal of Grid Computing, Vol. 7, Issue 1, Mar. 2009.
• Zheng, Q. and B. Veeravalli, (2009).“On the Design of Communication-aware Fault-tolerant Scheduling Algorithms for Precedence Constrained Tasks in Grid Computing Systems with Dedicated Communication Devices," Journal of Parallel and Distributed Computing, Elsevier Science, Vol. 69, Issue 3, March 2009.
• Li, X., Iyer, S., and Palit, H.(2009).“Design of a Model-based SLA Management Service and its Applications over a Shared Services Platform”, Grid Computing: Technology, Service and Application published by CRC Press, a Taylor & Francis Company, 2009.

Patents:

• Sharad Singhal and Yinping Yang (inventors listed in alphabetical order), “System and method for negotiating a sale”, US patent filed on 29 December, 2009, application number 12/648,405.

Awards:

• Cheng, Z., Yang, Y.P., and Lim, J. (2009). “Cyber Migration: An Empirical Investigation on Factors that Affect Users" Switch Intention in Social Networking Sites”, In Proceedings of the 42nd Hawaii International Conferences on System Sciences: HICSS42, Track 7, Big Island, Hawaii, January 5-8, 2009. (Best Paper Award).
• Yang, Y., Singhal, S., Santosa, R. Yan, Y., Foo, Y. (2009). “Negotiation as a Service: An Agent-Based Negotiation Engine for Online Business”, 19th Annual Workshop on Information Technology and Systems (WITS2009), December 14-15, 2009, Phoenix, USA. p. 224.(Best Prototype Award)

Collaborators:

• Hewlett-Packard Labs, Infocomm Development Authority of Singapore (IDA), Pacific Rim Applications and Grid Middleware Assembly (PRAGMA), Institute for Infocomm Research (I2R), Data Storage Institute (DSI), National University of Singapore (NUS), Nanyang Technological University (NTU)

Cardiac Function Modelling

Modelling of a human organ is especially a more difficult task than modelling a physical industrial object, due to the irregularities of its surfaces and its constant motion. In DMV, we reconstruct 3D models of human hearts throughout one entire cardiac cycle. Using the time-series 3D models, we extract indices that are computed from the curvedness of the heart’s surface. These indices are able to provide medical doctors with an overview of the health condition of the patient’s heart. Our team is currently using this new knowledge to perform actual test bedding in hospitals in order to validate the clinical viability.

4D Heart Modelling

Motion of the heart’s surface can be used to simulate and reconstruct the nature of the blood flow that occurs within a patient. Using the 3D heart models that we have reconstructed, our team plans to model the blood flow within the patient’s left ventricle using CFD simulation techniques. In addition, we are currently modelling the mitral valve that is located between the left atrium and left ventricle. The mitral valve controls the blood flow between the left atrium and ventricle. By simulating the function of the mitral valve and the blood flow that occurs, we believe that the data will be very useful for medical doctors in performing diagnosis on the patient.

4D model of left heart viewed from the apex

Lateral view of left heart

Scientific Visualisation

Computational simulation generates massive amount of numerical data which is difficult to interpret directly. Scientific visualisation is the post-simulation process of representing such data in an intuitive graphical format. By leveraging human visual perception, it allows researchers to derive understanding from the simulation results. This is especially important as the amount of data is growing with the increased in simulation complexity.

The DMV team manages the 2x3 tiled display in IHPC. The system is powered by the HP Scalable Visualisation Array (SVA) and provides a high resolution display of 4200 x 2100 pixels for visualising scientific data. This provides about eight times the resolution of a typical computer screen. The tiled display also has stereoscopic display capability. With the use of active shuttle glasses, scientists are able to visualise complex spatial data with depth perception. Such stereo-enabled viewing further improves the interpretation and understanding of data.

Representative Publications:

• S.Y. Yeo, Liang Zhong, Yi Su, Ru-San Tan, and Dhanjoo N. Ghista. ”A curvature-based approach for left ventricular shape analysis from cardiac magnetic resonance imaging”. Med Biol Eng Comput 47(3): pp 313-322, Mar 2009.
• Liang Zhong, Yi Su, S.Y. Yeo, Ru-San Tan, Dhanjoo N. Ghista, and Ghassan Kassab. “Left ventricular regional wall curvedness and wall stress assessment in patients with ischemic dilated cardiomyopathy”. Am J Physiol Heart Circ Physiol 296: H573-H584, 2009.
• Li Zhang, Yu Zhang, Chew Lim Tan, “An Improved Physically-based Method for Geometric Restoration of Distorted Document Images”, accepted for publication in IEEE Transactions on Pattern Analysis and Machine Intelligence, 30(4), pp 728-734 (Apr 08).
• Yu Zhang and Norman I. Badler, “Face modeling and editing with statistical local feature control models”, International Journal of Imaging Systems and Technology, 17(6), pp 341-358 (Apr 08).
• C.S. Chong, A. Senthil Kumar and H.P. Lee, “Automated Mesh-Healing Technique for Model Repair and Finite Element Model Generation”, Finite Elements in Analysis and Design, 43(15), pp 1109-1119 (Nov 07).
• Yi Su, A Senthil Kumar, “Templatized Refinement of Triangle Meshes using Surface Interpolation”, International Journal for Numerical Methods in Engineering, 65(9), pp 1472-1494 (Feb 06).
• Yi Su, KH Lee, A Senthil Kumar, “Automatic Hexahedral Mesh Generation using a New Grid-based method with Geometry and Mesh Transformation”, Computer Methods in Applied Mechanics and Engineering, 194(39-41), pp 4071-4096 (Oct 05).
• W.S. Li, S.H. Xu, G. Zhao, L.P. Goh, “Adaptive Knot Placement in B-Spline Curve Approximation”, Computer-Aided Design (Special Issue), 37(8), pp 791-797 (Jul 05).

The research agenda is as follows:

• Accelerate performance of algorithms/dwarfs by mapping them efficiently onto hybrid architectures such as Multicore, GPU, FPGA; where a dwarf is a function with certain pattern of computation and communication.
• Create performance meta-models of algorithms/dwarfs. With these models, efficient scheduling decisions can be made so that the most appropriate hybrid architectures are used for each algorithm mapped.
• Schedule algorithms/dwarfs efficiently onto hybrid architectures for both serial and concurrent scenarios, based on the performance meta-models created. Scheduling is done at several levels: sub-system (i.e. within a multicore processor), system (i.e. within a compute node), and trans-system (i.e. a cluster of compute nodes).

In addition to research, this joint lab has the following objectives:

• To serve as a platform for encouraging active interdisciplinary collaboration between the two institutions in the frontier research of high performance computing.
• Nurture Science & Engineering talents through manpower training for undergraduate and graduate students, and post-doctoral researchers, in the area of large-scale high performance computing for the scientific communities and industries.
• To encourage mutual participation and organisation of conferences, seminars and courses.

Academic Visitors:

Professor James C. Browne, The University of Texas at Austin, USA.