Thursday, 1st November 2012, 3:00 pm (PDCC Meeting Room)
Speaker: Mixed (PDCC)
Title: PDCC Group Seminar with BBQ
Speaker: Debraj Roy
Title: An integrated agent based simulation framework to explore slum formation.
There is a growing amount of work in the field of ABM and urbanization. However, the use of ABM to understand slum formation is still new. Agent Based Model demonstrates the behavior of individual agents based on their interactions with other agents as well as the environment. This can be used to understand the expansion of global phenomenon such as slums. The initial idea is to create and Agent Based Model (ABM) integrated with Discrete Event Simulation (DES). DES will be used to simulate population dynamics based on empirical data while ABM will be used to simulate human behavior and housing dynamics. The model will use a bottom-up approach. The design will be aimed at exploring the relationship between spatial pattern of the city and location of slum formation.
The model would be a useful tool to assist the policy makers in making robust decisions. The research also aims to answer several “What-if” questions regarding slum formation. The research will provide a simulation environment to conduct experiments pertaining to slum policies.
Speaker: Jianlong Zhong
Title: Kernelet: High-Throughput GPU Kernel Execution with Dynamic Slicing and Scheduling
Short Abstract: GPUs have become an effective accelerator for high performance computing. GPUs are usually shared by multiple users, and many kernels are submitted to the same GPU for executions. Throughput is an important metric for performance and total ownership cost in such computing environments. However, current multiple kernel executions can severely underutilize the GPU, resulting in suboptimal throughput. In this work, we propose Kernelet, a runtime system with dynamic slicing and scheduling techniques to improve the GPU utilization. Kernelet takes multiple kernels as input, dynamically performs slicing on the kernels. Our experimental results demonstrate up to 40% performance improvement over the recent approach on consolidated multiple kernel executions.
Speaker: Zhou Pengfei
Title: IODetector: A Generic Service for Indoor Outdoor Detection.
Abstract:The location and context switching, especially the indoor/outdoor switching, provides essential and primitive information for upper layer mobile applications. In this paper, we present IODetector: a lightweight sensing service which runs on the mobile phone and detects the indoor/outdoor environment in a fast, accurate, and efficient manner. Constrained by the energy budget, IODetector leverages primarily lightweight sensing resources including light sensors, magnetism sensors, cell tower signals, etc. For universal applicability,
IODetector assumes no prior knowledge (e.g., fingerprints) of the environment and uses only on-board sensors common to mainstream mobile phones. Being a generic and lightweight service component, IODetector greatly benefits many location-based and context-aware applications.
Speaker: Kyumars Sheykh Esmaili
CORE: Cross-Object REdundancy
Replication --as the most basic form of redundancy-- has been widely used by storage systems to assure availability. However, now that the storage providers are reaching the scale of petabytes, this form of redundancy can be prohibitive. As an alternative, Erasure Codes have been increasingly embraced as the preferred redundancy mechanism, because of their low storage overhead and high fault tolerance.
One essential aspect of Erasure Coding is the repair efficiency. Currently, there are a number of proposals on
designing Erasure Codes with better inherent repair capabilities. In this work, we adopt a different approach by
proposing to code across different already encoded objects to alleviate the repair problem.
A number of speakers all from PDCC will spend 25-30 minutes outlining their current research.